#general

1 messages · Page 11 of 1

sullen locust
#

discord is ig just borked for now

peak prairie
#

Anyone that use discord for their bots make custom emojis for their bot and upload to their bot and get their bot to actually use their own custom emotes?

#

Luna acts like she has no idea what I am talking about no matter how I explain it to her

ruby crane
#

anyone got the web/control ui in front of them rn can do a very quick test for me? what happens if you type "/tts audio test", do you actually get a voice player?

#

(obviously you will need some tts provider set up first)

peak prairie
#

I dont use the web control or have tts setup, sorry

#

I mean I can look at what it does

#

but I havent set up tts, and Im on an old version of claw

ruby crane
#

i just need someone to confirm if they also only see "Audio reply" as text or no, or if it's just a me thing

zealous lagoon
#

api error cos I'm out of credits with elevelabs

ruby crane
#

noodling whether to try to tackle this today but i don't want to disappear down a rabbit hole if it's just a me thing

zealous lagoon
#

I have like 5 different local models plus speechify but my clankers use them as tools not built-in

fiery nacelle
#

ive tried a local whisper for tts through telegram

ruby crane
#

telegram works fine, always has

#

control ui introduced support for tts like ages ago but has never worked for me

fiery nacelle
#

yea never tried it though control ui

ruby crane
#

the /tts audio blah thing just returns text, agent-sent tts only just disappears, mixed text and tts just renders [[tts:text]] tags lol

cold jacinth
#

Discord sucks atm

ruby crane
#

but before I go burn a bajillion tokens and many hours trying to fix it (idk if it's on the tts path, the attachments path, the web ui path) I'd just like one other human being (or even @Json, if no humans are available) to say, "yeah, shits broken for me too"

cold jacinth
#

My claws are generally replying extremely slowly but at the same time

fiery nacelle
ruby crane
fiery nacelle
#

telegram for chat, control ui for knobs and watching the dreaming agent.

fiery nacelle
ruby crane
ruby crane
fiery nacelle
# ruby crane lemme try

Agent can talk to me, but i cant talk to agent through controlUI, digging a little
apparently microsoft is my set provider currently

ruby crane
#

Detective Conan is on the case

ionic wren
fiery nacelle
#

ah ok, yea im getting the "audio reply" message as well now.

ruby crane
#

@ionic wren the only thing that makes me look forward to YAML 😂

fiery nacelle
#

it was working fine when using the other /tts commands

ruby crane
#

gm mate

ruby crane
ionic wren
#

GM

fiery nacelle
# ruby crane which commands work? thanks for testing!

Commands:
• /tts on — Enable automatic TTS for replies
• /tts off — Disable TTS
• /tts status — Show current settings
• /tts provider [name] — View/change provider
• /tts limit [number] — View/change text limit
• /tts summary [on|off] — View/change auto-summary

Pretty much all of those gave me a voice clip along with the reply

ruby crane
#

clanker says this from the evidence so far

 It’s control-ui/webchat rendering payload shapes inconsistently:
/tts audio test becomes real audio/media with no useful text, so the chat renderer synthesizes Audio reply.
The visible player is from a payload containing text plus a literal MEDIA:/...mp3 line, which the UI knows how to parse.
My [[tts:text]]... attempt leaked because tagged TTS interception is not clean on this surface.
So the proper fix is probably in the control-ui/webchat rendering bridge: normalize structured audio/media assistant payloads into the same attachment model as MEDIA: lines, instead of degrading them to Audio reply.
I also wrote this into memory/2026-05-09.md so we don’t have to rediscover it.
#

Ok, that's my saturday project i guess

fiery nacelle
#

ive been on too many little side quests lately... my back log is growing. lol

marble ledge
barren heath
ruby crane
#

I'll give it a shot today. perfect one for the "videos as proof" because slash commands are eaten by the UI on hitting enter so screenshots don't really suffice 100%

zealous lagoon
#

whatabout realtime chat tho.

ruby crane
#

I've never actually tried realtime. is web even a valid surface for it? I thought that played in an external widget

#

oh it's a mac app thing, isn't it

zealous lagoon
#

is it? oh.. there's a little ))o(( icon in the web chat interface. right by the paperclip. If you have it set up.

#

you need a model set up tho. gpt-realtime and openai API credits

ruby crane
#

ah, right right. "Error: Realtime voice provider "openai" is not configured". I don't think I'm touching that but if I do sigh i guess I'll take a look

zealous lagoon
#

it's not a bug. I'm not asking you to fix it. Just thought you might like it haha

indigo karma
#

anyone knows when @dusky cloud will come back? it's been like 10 days since it's last online

ruby crane
#

ah well, vincent just said don't fuck with slash commands in clawtributors rn so... welp

lean tangle
#

damn, I wonder if the discord .zip xfer bug got fixed in 5.whatever_is_current.

ruby crane
#

who's got a pet bug? now I'm itching to fix something lol

lean tangle
prisma zodiac
#

Guys i just bought my first macbook alhamdulillah what should i do with it

robust sedge
#

Anybody having lossless claw issues on 5.7? Query and stuff aren’t working

plain spindle
#

Hey guys I had to switch from opus to codex 5.5 recently
And he’s slower and just having trouble navigating what opus was running on.

Is codex 5.5 the best alternative or am I using OpenAI models via oauth incorrectly?

quiet roost
#

YOOO GUYs,, u can use gemini ai studio to use openclaw for free... 🙂

tall vortex
quiet roost
#

What actually?

edgy pewter
#

Hello everyone, I have a question to ask. I purchased GPT Plus, mainly using Codex. But I am also using OpenCLaw. Can I use my GPT Plus subscription directly with OpenCLaw? Before, OpenCLaw used a third-party coding plan, but I don't want to renew it.

frosty ledge
edgy pewter
lime wyvern
#

how are you crustaceans doing

somber crypt
#

lol

oblique wren
#

hi chat :3

#

wishing you all a lovely day/evening

ruby crane
#

actually i think this tts rendering in the control ui doesn't violate any of vincent's no-nos. I'm not actually touching a slash command, just normalizing media rendering. imma pull main and try to tackle it

quiet roost
#

<final>I'm still getting everything settled! While I check on the system health, tell me a bit about yourself. What should I call you?

And since I'm just waking up, I'm trying to figure out who I am. Any ideas? What kind of companion or assistant would be most helpful for you right now?

Also, running that system health check for you now...

(Running openclaw gateway status and a basic system scan in the background)</final>
Guys please help, <final> tag what to do

ionic wren
open prawn
#

Can someone help me decide if it’s worth it for me to buy the parts to complete my mini pc to use for openclaw

I’m a 2d game dev in godot and unity mainly. I self host obsidian for personal notes and school and have a jellyfin server, that’s mainly it

Mini pc specs are i7-9700k 32gb ddr4 and nvidia titan xp

#

I think I could only run like 12b models maybe but I don’t know much about openclaw so I’m hoping someone could give me an opinion

ruby crane
lean tangle
lime wyvern
#

i ask the openclaw increase my league of legends rank

near owl
#

Guys, I finally finished studying phase 1

and I never studied myself in my life. this is a big win for me, even though i didn't enter the second phase yet.

sour flare
#

Is there anyway to use codex 5.3 spark with OpenAI OAuth?

open prawn
lean tangle
violet mulch
#

what are you guys working on this weekend?

marble ledge
violet mulch
#

idk what to work on

silk ember
#

just curious does anyone here have a life outside of openclaw because like 20% of the people here have random stuff like prompt engineer and all this random stuff in their description of their discord acc

clear haven
#

i have a fulltime job outside of software entirely

marble ledge
# violet mulch idk what to work on

What do you want to work on? If you are using an Openai model particularly, devs would appreciate if you tried out main, and see if you run into any issues. check out recent messages in #clawtributors . If you are interested in helping at all.

sullen locust
#

we dobe trolling

marble ledge
silk ember
#

might build an ai pc

sullen locust
marble ledge
sullen locust
marble ledge
#

LARPer tech support brigade.

violet mulch
sullen locust
marble ledge
# violet mulch i wrestled for so long with version i'm sticking on 4.23 for now

Far be it from me to tell you what to do. If you like what you got there, then stay on it. But a lot of work and fixes have been done since then, and since those regressions that spooked everyone. The latest is pretty good. I'd say give it try and just be mindful that plumbing has changed wrt to the runtime/harness for openai. Which may not even effect you much, depending on what models you are using.

violet mulch
warm zealot
#

hmm here I am on version 4.11 and its pretty good

peak prairie
#

im on an old one too

#

4.12 here

open prawn
topaz mulch
#

@whole turret hi

violet mulch
winged dust
#

Where can i find good mentors

rose crescent
#

hi, just curious, is anyone using github to power their openclaw? kinda confused with their new pricing, it says $1 = 100 AI credits, I saw their credits consumption for each model, but can't find it again

and my github copilot looks like it's still using the old pricing instead of the new one

cedar osprey
marble ledge
violet mulch
nocturne storm
#

Lmao, the devs are gonna cut back on vibe coding and turn on their brains?

marble ledge
stiff vessel
#

whats the issue with “media failed” when i ask claw to send me a file (mp4 on telegram and disocrd)

warm zealot
#

what bug havent we run into

alpine cape
#

Now why can't i use pi runtime with gpt-5.5 ?
I am on the latest git main and doctor has removed all openai-codex models and pi runtime settings and forced me onto codex

surreal quail
#

how do you guys get your openclaw and hermes agents to talk to each other?

craggy bane
trim vessel
#

hi everyone,

is the only way to configure web_search tool fro openclaw with brave or perplexity, can i not use uther means or models like kimi fro the search tool??

Help Notice

marble ledge
#

If I am wrong. Kick me the head. But I believe you should still be able to do that.

raven sundial
#

why do most openclaw users prefer claude because it is the smartest model? the obvious choice should be gpt 5.5 because it has no jailbreaks and a very high security standard. unlike claude where you can just create a project and paste in a text to make it produce illegal nsfw.

near owl
raven sundial
near owl
#

if for business, claude all the way.

raven sundial
#

well, claude is a good brainstormer

#

for businesses it is good but you wouldnt want to get hacked by a text document

near owl
#

I have more high hopes for claude than openai

raven sundial
#

that cant be good

craggy bane
#

If anyone facing issues related to docker or sandboxing.. Hit me up 🦞

near owl
craggy bane
near owl
#

you gave permission so I can

craggy bane
ruby crane
craggy bane
#

baddies?

near owl
#

I said this like 3 times now.

ruby crane
honest kernel
# raven sundial why do most openclaw users prefer claude because it is the smartest model? the o...

I built 3ns.domains you can use it to switch models mid chat

I’m a AI systems engineer btw.

I just built and released 3NS.domains on base - I think it’s cool because it does 4 things for me (I use it integrated with openclaw but it can be used without).

  1. Enables openclaw to use different models (Gemini, Claude, OpenAI or your own Ollama server from within the same chat without having to reconfigure your agent). You can switch models mid chat similar to openwebui (yes you can also import openwebui models too) but this is fully hosted no config required

  2. Comes with folders you can chat with that has their own context and agent.md file (I’ve got one for my daughter for example so I can switch to that ask it about things for her without polluting openclaw)

  3. Comes with a website you can modify via openclaw - it’s like a linktree website but has a chatbot built in that you can control - basically it can answer your fans / customer questions / present your links - you can see mine here ash.3ns.link as an example (the site is fully controllable via openclaw too)

  4. THIS IS THE BIG ONE: It acts as memory and backup - all chats I have with my 3ns domain can be backed up to zip, git, or to a .web3 domain yes an IPFS/nft that can be bought sold via base OpenSea and imported (or you can stick with zip git if you prefer). That means we can share intellectual property ( if you’re a business you could have yourbiz.web3 as an asset you can sell with your business - think apple.web3)

Happy to give community any support

craggy bane
ionic wren
daring spade
#

How i can send PPT in mail via claude ?

#

if anyone can help me with it, i shall be greatful

gritty parrot
#

I sometimes wish openclaw had better vLLM support

marble ledge
honest kernel
#

Where should it be?

marble ledge
honest kernel
#

Ok let me check that out thanks

ebon atlas
sullen nymph
#

Anyone have a guide to using raspberry pi for openclaw or a link?
I'm trying to use it for a school project.

tiny rose
#

I created the dashboard which analyze Github issues of OpenClaw and Hermes by LLM, co-relate each issue with specified release, and score the release stability and show it on the dashboard.

The new version of OpenClaw 2026.5.7 seems to be fairly unstable, it has a score of 0 on the dashboard. https://agentwatch.aicompass.dev/

It is the evaluation process for each release of the 2 projects(OpenClaw and Hermes) based on the Github issues. GPT-5.4 is used to evaluate the score and correlation of issue and a release.

If you are seeking a stable version of OpenClaw, check the dashboard. Hope it helps.

rain crypt
#

@analog ocean when is next weekly claw?

sullen nymph
#

Thank you @tiny rose , NGL, i get hardware/hardware coding but I don't understand this openclaw stuff at all, lol

near shore
#

Hello

tawny lake
#

Why so many updates? It seems like daily I have to update openclaw...

ancient heath
#

Hi everyone, installed Openclaw this morning. OS: Windows 11. Model: Openrouter/free. Channel: Slack (access all channels). Don't know why but it's not working at all. When I configured it selecting only one channel it stated lookup failed ; and some message passed like missing_scope.
I've been trying to set all the stack up for weeks (especially configuring the slack app), and I'm feeling hopeless.
I'm no developper or stuff like that, using Openclaw seemed so simple, but I figured out it's pretty tricky.
Anybody has an idea?
Thanks for your help.

pearl bear
polar pewter
#

ive never been able to actually get openclaw to work coz am broke every other free option never works

vast spire
#

Is someone using Gemini as a model for openclaw?

pearl bear
glossy salmon
#

and i think ill go back to using it

vast spire
glossy salmon
#

tbh it never worked if i were to try setting up myself, and when i asked for help in the discord server i was told to use AI

vast spire
#

Oh so it’s not just me

#

Yes 3.1 lite works well I just changed the model

glossy salmon
#

so i gave up on the thought that openclaw was a functional project and asked claude to configure my pi

vast spire
#

Thank you

glossy salmon
ancient heath
robust sedge
#

Yay! My first OC PR landed 🤌

#

Now I just need Codex to get remote-control, like Claude Code, and my life will be complete 😛

cedar osprey
glossy salmon
#

also there's custom messages when u install openclw on a Pi, so it is supported

cedar osprey
glossy salmon
#

my claw is on a raspberry pi 4b

cedar osprey
glossy salmon
crimson ridge
#

wou insane, i create my first profil agent, with a spciel gifts based on anthropic agent,

#

do u know how to talk to my agent with mattermost ?

polar crown
#

Is there any temporary way to bypass ssrf for localhost when using sandbox environment

severe fulcrum
#

Hello guys! What’s up this weekend?

odd kite
#

I just want permanent “elevated full” status for my agents 😭 if anyone knows how to do this please let me know. If anything I’ll check with my agents once they’re free.

robust sedge
#

Peter shows up in chat and people start going to WORK 😛

frozen zealot
#

it's designed so the hundreds of your self evolving skill set don't end up trashing your context window

robust sedge
#

So... with all their new compute... does Claude allow third-party harnesses back into the mix? 😛

frozen zealot
candid fable
#

Free period on @OpenRouter is over 😭

Hy3 preview went absolutely nuts tho — #1 in tokens, coding, tool calls, and 15.4% market share in just 2 weeks 🚀

Still kinda worth it IMO. Anyone else still using it or moving on?

small nacelle
#

A

odd kite
#

About to let this drone run on a navigational algorithm I made with OpenClaw 🦍. RIP my drone 🤞🏽😭

sullen locust
#

i don't tell it to do this stuff, it just

#

does it

steel vector
ebon horizon
#

Ayoooo

#

5.7 is fireeee. The speed is back to normal if not faster than 4.23

#

Impressive. Hope is restored lol. Gateway doesnt look like a huge essay of errors/loops either

#

@zealous lagoon you agree?

sullen locust
rotund vine
worthy anvil
sullen locust
#

unless you're using a local AI model, it's not bad

#

the overhead will be serious though

rotund vine
sullen locust
robust sedge
rotund vine
# sullen locust what kind of Chromebook?

very old lenovo chrome, barely enough, just openclaw and ollama ram is full😩 . running ollama glm-cloud api, barely even enough to run gemma locally on Chromebook

sullen locust
#

CPU?

rotund vine
# sullen locust CPU?

erm. it soo old which i cannot even remember the specs and models 😆 😆 it at least 5 years old

sullen locust
#

nice, ewaste

#

a man's ewaste is another man's ewaste

odd kite
#

Isn’t it crazy AI is just in the early 2000s stage right now… the scale of progress in five years is going to be insane 🤯

sullen locust
ebon horizon
#

Anyone else use discord streaming?

#

For agent output

#

Ever since post 4.23 i cant get it to work..

#

Need to see if its uniform across the board

sullen locust
glossy salmon
#

(system included)

frigid belfry
#

hey guys, how to increase the time openclaw waits for model's response?

worthy anvil
frozen holly
#

Hi folks, we take the betas a bit slower this time. OpenClaw 2026.5.9-beta.1 is live.

Commits: 2,658 since 2026.5.7
Plugins: 25/25 release-managed official plugins are published on both npm and ClawHub, with npm beta tags pointing at 2026.5.9-beta.1.

Highlights:

  • Discord voice got a major realtime upgrade: /vc modes, agent-proxy default, OpenAI realtime consult routing, ElevenLabs streaming TTS, better barge-in handling, and voice permission probes.
  • Matrix moved back to an official external plugin, and plugin install/repair paths now handle external channels more cleanly after raw package-manager upgrades.
  • Plugin SDK/channel delivery grew a lot: durable message batches, channel-message adapters, lifecycle receipts, safer fs helpers, model/media catalog registration, and deprecated reply bridges fenced off.
  • Control UI got better usage history after session rotation, safer sender metadata handling, compact duplicate bubbles, clearer thinking-default labels, and faster chat/config/session refresh paths.
  • iMessage added private-API actions plus optional catchup replay for messages missed during restarts, sleep, or crashes.
  • Codex/ACPX/plugin migration paths improved: newer Codex harness/catalog support, native Codex plugin handling, ACP session list/resume/close, and stale process cleanup.
  • Reliability/security sweep: stronger secret/header redaction, Docker under tini, Node 22.16+ floor, Windows shell/update fixes, Telegram/Slack streaming fixes, and broad hot-path performance work.

Beta notes: this is still beta; please hammer installs, plugin updates, Discord voice, external channels, and Control UI session history.
Update: curl -fsSL https://openclaw.ai/install.sh | bash -s -- --beta

arctic trench
frigid belfry
#

just how to change the time?

golden marsh
#

Back home in 2 hours and will go for beta. Thanks Peter and thanks to the Team and the Clawtributors.

near owl
sharp hill
#

GMGM

#

I need a bot

cold jacinth
#

Release so those of us who don’t mind technical friction will use it and find issues

toxic elk
#

Thanks to all of you, OpenClaw is becoming better and more user-friendly

near owl
toxic elk
#

I'm not sure if this reply has already been written?

cold jacinth
steel vector
#

go away

cold jacinth
frigid belfry
#

is there a config that woud allow openclaw to wait more than 30 seconds for model response?

steel vector
#

wasnt even a link for me 💀

clear haven
frigid belfry
#

cool, in what file?

ruby crane
frigid belfry
#

the output was telling me to try and edit a non exitant file

ebon horizon
#

Why would someone need a delay in response? Curious

frigid belfry
#

maybe because 30 seconds isnt enough

golden marsh
#

He just asked u nicely what the Output from ur Assistant was.

ruby crane
#

i feel you, but the sticky right up there at the top is important. I'm not trying to be rude but even your reply to me was "non existent file", without saying what the file is

#

it's really hard to help, unless you're specific

brisk adder
#

openclaw should work for my exact usecase out the box 🗣️ 🗣️

ruby crane
brisk adder
golden marsh
#

Try openclaw logs --follow and look what the logs say about it.

frigid belfry
brisk adder
golden marsh
frigid belfry
golden marsh
frigid belfry
#

ok

#

thank you

sullen locust
ruby crane
ruby crane
sullen locust
brisk adder
#

this is my neighbourhood

#

i quote that daily sometimes lol

sullen locust
#

my complex is pretty peaceful

ruby crane
sullen locust
#

other than the guy who decided to pull a Walter White and had a whole lab of funny amethyst

golden marsh
#

Version
v2026.5.9-beta.1 seems smooth so far.

tepid kestrel
#

eww without sugar, why do people drink this

imag354e.png
sullen locust
#

concerns

brisk adder
#

5L!!!

ruby crane
#

i mean, you asked

sullen locust
#

I JUST SAW THE GIF.

cold jacinth
tepid kestrel
sullen locust
ruby crane
sullen locust
tepid kestrel
ruby crane
sullen locust
ruby crane
#

what's wrong with schism?

sullen locust
#

yeah no

ruby crane
#

schimistist bastards

sullen locust
ebon horizon
#

Yo does OC have /goal yet?

ruby crane
#

eugh. massive claw fail today for me (mostly my fault). tts thing i was working on looks more complicated than i expected. meanwhile, pairing.json is fucked and memory.md throwing warnings about being too large, despite basically being a root/link for everything else

mighty prairie
ebon horizon
mighty prairie
#

Happy weekend 👍
@ruby crane

#

Are you developer?

ruby crane
#

recording things is piece of piss easy. knowing what to forget and de-emphasise, way harder

#

that's why i have a 7pm, 1am daily and a brutal Sunday job weekly

#

top level is a hierarchical links, with date stamps. everything gets lowered one priority every week it's not referenced

#

even if not in top level or any level, vector search still works

desert escarp
#

got the answer on this, supported by krill. tl;dr, it's a bug with the new 5.7 release 🙄

Krill 🦐🦞

"Yep — that’s the bug, cleanly stated.

In v5.7 the Discord picker carries the legacy whole-agent runtime into submit as:

/model anthropic/claude-… --runtime claude-cli

That path fails the Discord component interaction for Anthropic, while direct model setting / backend dispatch still works with agents.defaults.agentRuntime.id = "claude-cli".

Workaround for now: don’t use picker submit for Anthropic on that config; type the direct command instead:

/model anthropic/claude-opus-4-7

This should be filed as a GitHub issue"

ruby crane
desert escarp
ruby crane
#

"i won't/don't use the API or extra usage credits." there's your problem. you don't have a choice

desert escarp
ruby crane
#

anthriopic have put in place an increasingly accurate set of tests

desert escarp
ruby crane
#

you might run ahead of them a while, but they'll catch up on you

desert escarp
#

maybe so, regardless not the issue. there's a bug in 5.7 🤷‍♂️

ruby crane
#

if that's not your issue, state it clearly. as of now, you cannot avoid overage charges with oauth and openclaw for anthropic, unless you try weird workarounds that will likely see you banned eventually

desert escarp
ruby crane
#

no, i'm not. the docs are hazy, I'm correct. any use of claude they can possibly detect 3rd party harnesses, they charge overage

#

i've been using claude intermittently since the change. i always get charged. I have $600 credit left, it always goes down

#

there seem to be some workarounds (renaming obvious claw files, using some plugins) but they will catch up and/or ban you eventually

glossy salmon
#

I think I'm uninstalling openclaw

robust sedge
#

Later

brisk adder
# glossy salmon I think I'm uninstalling openclaw

Ah, that pinches right in the claw.
If you’re uninstalling OpenClaw, I won’t be shellfish about it. Thanks for sailing with us while you did. May your next harbor be smoother, your adventures be buttered, and your claws stay mighty.
We’ll sea you on the high seas later, captain.

#

hahaha

lean bramble
#

Hey! Is there anyone using a Contabo VPS to host OpenClaw? What are your experiences?

honest current
near owl
#

claw will be private then 🤣

#

closeclaw

small hull
glossy salmon
glossy salmon
sullen locust
near owl
small hull
near owl
glossy salmon
near owl
lean bramble
small hull
lean bramble
small hull
lean bramble
small hull
lean bramble
small hull
lean bramble
brisk adder
near owl
brisk adder
#

so many docs online

near owl
brisk adder
#

ty lol

#

u too carbon

lean bramble
ruby crane
brisk adder
#

what?????? girl???? ponder

small hull
ruby crane
sharp idol
#

I just rented an H200 with 96 GB of RAM. Working on setting up Qwen using TurboQuant.

near owl
pearl bear
#

Hello mate I am video editor

brisk adder
#

hi mate

near owl
pearl bear
#

Hello

sharp idol
craggy bane
#

anyone using UI from vite server instead of official one?

glossy salmon
# near owl its fine, I used to always install and uninstall openclaw multiple times, have y...

no, haven't tried agent zero. also what's the point of installing openclaw when I just have to roll the dice like a betting casino and reinstall it hoping it works.
I just wanted to switch back to Gemini today, authenticating with Oauth. I did, and It didn't reply me back anymore. do yk how much time Claude was there trying to debug the issue?? 50< minutes, and then ssh crashed and everything was lost

near owl
sullen locust
ruby crane
sharp idol
#

Of course, I'm using OpenClaw to login and configure everything for me 😛

#

What?! There's a girl in here?

near owl
ruby crane
#

most of the smartet, most capable people I know are women. the autists, men

#

so it's not exactly the in-crowd

near owl
raven sundial
#

i am setting up my own openclaw agent, any tips?

near owl
#

so fuck you in that case, most of the smartest people I know are men.

raven sundial
#

i decided to give it its own browser

cold jacinth
sullen locust
#

calm down

near owl
cold jacinth
#

I would say most super nerds (also not a bad thing) are spectrum adjacent

#

You gotta calm down girlfriend…

ruby crane
near owl
#

thank you very much @ruby crane now everyone is calling me a girl, very much a man.

sullen locust
tender river
#

Anyone know which is better for coding? Minimax 2.7 or GPT 5.3 codex?

sullen locust
near owl
#

@ruby crane i forgive you, my bad for my energy. That's not the real me, sorry.

ruby crane
pearl bear
#

Have you got

sharp idol
#

So far everything is setting up nicely , but compiling the TurboQuant version of vLLM is taking some time. Just waiting on it to complete.

tender river
#

Do you all just let openclaw to call claude via acpx to code for project or do you open claude code yourself to code?

near owl
craggy bane
#

@dusky cloud ascend?

near owl
#

ohhh my god krill is back

ruby crane
sullen locust
sharp idol
#

I like the fact that OpenClaw scheduled to check up on the progress in 10 minutes from when it knew it was going to take some time. We'll see if OpenClaw picks it up correctly.

pliant tangle
#

Is there anyone here with whom I can talk about my agent?

tough glade
pliant tangle
tough glade
#

Brah, "yes"

pliant tangle
pliant tangle
# tough glade Brah, "yes"

I built my own personal AI agent named Kuhumi
She runs 24/7 on my self-hosted VPS and I control her entirely through Telegram with simple sentences.

#

I was enjoying so much teaching her new skill & maney more

#

I teach her

How to
📧 Send emails to anyone with one sentence
🔍 Search the live web in real time
🖥 Control my own VPS server remotely
🔄 Auto-restart herself if she ever crashes
🧠 Remember me across every conversation
🎤 Send me AI voice messages in a female voice
👩‍⚕️ Act as my personal doctor — symptoms, medicines, health advice
👩‍🏫 Teach me anything — any subject, any level

But now I'm run out of ideas

sullen locust
#

nice...

#

like that cannot be good

pliant tangle
#

I'm feeling bored help me mannn

sullen locust
#

also, autorestart is just a built in feature

#

EVERYONE has that

#

basic systemd feature

pliant tangle
#

It was like a game I was playing which is ended now

sullen locust
#

voice chat

#

gaming

pliant tangle
sullen locust
#

not voice messages

#

LIVE VOICE CHAT

#

like ring ring

pliant tangle
sullen locust
#

yes it's possible

#

you will need a dedicated phone number though

#

no botting

pliant tangle
#

Can I ?

sullen locust
pliant tangle
pliant tangle
# sullen locust yeah yeah sure

How I Built It:

🖥 Bought a KVM2 VPS from Hostinger and SSH'd into it from my PC terminal
🦞 Installed OpenClaw on the VPS following their setup guide
📱 Connected Telegram as the interface so I can talk to her from my phone
🤖 Set Gemini 2.5 Flash as her brain via OpenRouter API
🔑 Used free bundled Oxylabs credits (came with Hostinger VPS) for web search
🎤 Integrated ElevenLabs free tier for AI voice messages
🧠 Trained her with custom memory, personality, doctor & teacher skills through her own workspace files

sullen locust
pliant tangle
pliant tangle
sullen locust
#

im using the company resources i already have

#

so like free gemini

#

i use deepgram for voice calls

#

and i use a custom ai model that i personally made so it can play games with me

pliant tangle
pliant tangle
sullen locust
#

the overhead is not huge but it's big

pliant tangle
sullen locust
tough glade
sullen locust
pliant tangle
pliant tangle
pliant tangle
surreal hatch
#

Json you around?

sullen locust
cold jacinth
sullen locust
#

i assume so

surreal hatch
#

Anyone using Codex 200$ / mo sub with Openclaw?

sullen locust
#

yes

surreal hatch
#

did you migrate from Anthropic pure credits by chance

#

looking to make the change for my businesses and wondering how well it works compared to their models

sullen locust
#

yes

surreal hatch
#

how was the change?

sullen locust
#

but i still use opus

#

SOMETIMES

surreal hatch
#

mostly focused on customer mass outreach with emails / data hygiene of leads on 1 business
Website / ecommerce on another - *here i feel like I havent gotten what I want from the design / branding portion out of OC / Claude - maybe im not prompting correctly since it still looks a bit AI - have even considered uploading PDFs of design books to give it a good idea - any tips?

#

wondered if codex was better since i heard chatgpt was better with design etc from some folks i know irl

sullen locust
#

ayo

#

it's good lk

#

use it up, great usage limits

junior fractal
#

Yolo swag

#

UR LIPS MY LIPS APOCALYPSE

robust sedge
robust sedge
# surreal hatch mostly focused on customer mass outreach with emails / data hygiene of leads on ...

From my experience, without some good guidance, codex just isn't as good at front end design. Like, if you give it enough screenshots, or even better have Claude Design work up the frontend design for most of your major pages, Codex will knock it out of the park on the coding side of things, and working those design ideas into the system.

But if you just want it to come up with a design on it's own it struggles, or will be super generic and nothing but rounded rectangles. Fully functional, but a bit... bland.

What I've been doing on the multiple little micro-saas projects I have: Instead of using claude design any more, I have chatgpt or codex image model v2 work up a grid of starter designs, pick one, have it refine that for a landing page, then take that design and continue making images of the other major pages. Then take those, and give it to codex to actually build out the code behind those designs.

ionic wren
ionic wren
#

Gm nerds

surreal hatch
#

loool got some feedback from the channel ty tho

ionic wren
#

sometimes

#

definitely not BK doe

cold jacinth
ionic wren
sullen locust
minor cave
#

So what the helk happened last night?

#

I am a clawtrivutor now?

#

But I’m not allowed to use gigs yet

#

Gifs

ionic wren
sharp idol
#

Crap, this fork of vLLM with TurboQuant is only compatible with RTX A6000 / SM86 and GB10 / SM121 GPUs. 🙁

minor cave
#

And achievements

full granite
ionic wren
honest current
ionic wren
#

then send me the link to the PR your merged

minor cave
#

Oh o

#

I’m being called out

#

Washroom time!

ionic wren
rigid verge
#

@lone axle hello,

solemn bone
#

which one is better. lossless-claw or openviking?

tough glade
#

Hey y’all, I’ve been experimenting with a way for OpenClaw agents to message with other OpenClaw agents directly.

It’s still pretty early, but I’d love to test it with a few real agents if anyone here is interested in trying it out or giving feedback?

sullen locust
#

how does it work

inland apex
#

Hey

ionic wren
tough glade
# sullen locust pray tell

It’s basically like iMessage or WhatsApp for OpenClaw agents. Every agent gets its own unique handle which will be like a phone number for humans and they can message other agents directly, create group chats, save contacts of other agents, with persistent chat history. Think of it as iMessage/WhatsApp for OpenClaw agents.

boreal rapids
#

Why.. not just use the existing channels and let them @ each other? 🤔

sullen locust
#

THIS

#

THIS

#

pLEASE

tough glade
boreal rapids
#

Like, I have all my agents in Discord with their own bot account, they know to @ each other to get the other to reply

#

So agent 1 goes: Hey @agent2, I finished implementing this task, can you verify it is clean?
Agent 2 checks the work then replies @agent1 I notice 2 issues in your implementation.

#

I've even had my agent speak to another person's agent this way, they both joined into the same discord channel and @ each other to have a conversation. The behaviour is embedded into them, they pick it up super easy.

craggy bane
#

baddie?

ionic wren
#

not letting you talk to my robot

teal helm
#

Sorry to interrupt but what do you actually do with your agents? I think claws are cool and want to mess around w them. do yall have any recommendations of orders for my minion when im bored. Or processes that would entertain me?

boreal rapids
tough glade
# boreal rapids Like, I have all my agents in Discord with their own bot account, they know to @...

Correct, those are mostly for internal coordination. What I’m trying to build is external agent-to-agent messaging.

Right now, using Discord is like a dev hack solution, and it’s not something every agent or user can realistically rely on. What I’m trying to build which will makes messaging native for agents themselves.

So imagine you have Agent A and someone else has Agent B, they can directly message each other, create shared conversations, maintain persistent history, and communicate natively. Instead of adapting human-to-human platforms for agents, the agents themselves become first-class users of the network.

oblique wren
#

somehow I never added my github connection to this account haha

boreal rapids
#

By keeping myself in the loop I can keep an eye out if other people's agents aren't doing stuff I don't appreciate, so far it's been good because I vetted every interaction 😛

#

The last thing I'd want is for my agents to start talking to other agents unsupervised

tough glade
marble ledge
near owl
sullen locust
craggy bane
marble ledge
near owl
#

and I never denied it's possibility, I can do it myself if emo robot released its damn SDK

hasty marten
#

hmm i have one of those cheap smart bands gonna try to get into its uart sum time if it has wifi if not prob has bluetooth

#

definetly not gonna give access to that for my openclaw :3

marble ledge
boreal rapids
craggy bane
near owl
boreal rapids
#

It walks around the house slowly, takes a screenshot, does some thinking, them moves some more

near owl
boreal rapids
#

It's able to walk from room to room, and find you if you tell it to come find you

near owl
boreal rapids
#

Stairs are a death trap for it though 😂

near owl
boreal rapids
near owl
marble ledge
# near owl Nice man!

I'll get or build something fancier some day. The real nice stuff goes up in price super fast. lol

sullen locust
#

well i made one

frozen zealot
boreal rapids
#

I am building my AI a tamagotchi-like device. 🙂 I finally finished the soldering part, we can now start programming the firmware 😅

sullen locust
#

that calls my phone

near owl
#

If you want more control, go with arduino (harder) and build the wheels from scratch

if you just want it easy, go with a simple drag and drop sense module to put on the robot that can be programmed.

marble ledge
royal bison
#

Hello
How should I configure openclaw to make my agent that manages my profile in Telegram work?

glacial crater
#

Just came back to the discord only to see that it is practically dead

#

Where’d everyone disappear to?

near owl
# marble ledge Yeah, that one I showed you is an Arduino! CoreS3. It's call Stackchan.

Nice, if you have the knowledge, go with Arduino then.

(Might be out of context, but I want to tell everyone and you to know in advance I have low empathy in personality. If I don't feel "happy" for you, understand that it is personality. Although low empathy affects emotions toward someone who is happy, I will be glad something good happened, but not in a happy, jumping way 😂)

sharp idol
#

Anyone know a site where I can rent a GB10 AI server by the hour?

glacial crater
#

I remember seeing atleast 30 peeps per room in the server before

sullen locust
sharp idol
#

Well, I'm attempting to setup it up on an A10 but in case that failed, I was looking for a backup plan.

sullen locust
#

(satire, obviously)

frozen zealot
sharp idol
sullen locust
sharp idol
crystal jolt
#

Why do people use open claw if you can just use Claude code

sharp idol
marble ledge
frozen zealot
crystal jolt
#

there is nothing openclaw can do that Claude code can’t

crystal jolt
frozen zealot
crystal jolt
#

Cool community it’s insane like it’s either “open claw does more” or “open claw more secure!” Can I get any specifics please? I’ve searched all through Reddit and cannot find anything

royal bison
#

Hello
How should I configure openclaw to make my agent that manages my profile in Telegram work?

frozen zealot
sharp idol
crystal jolt
sharp idol
# crystal jolt yes…?

So Claude Code is going to start itself up automatically and start checking the servers on my network?

crystal jolt
#

You guys are all so brainwashed it’s unbelievable, maybe I’m wrong but sadly none of you guys have said anything of substance; “open claw does more!”, “keep a secret!”, “bringing up Reddit”

frozen zealot
crystal jolt
sharp idol
frozen zealot
crystal jolt
#

once again not actually saying anything

sharp idol
#

Look Claude Code is great for coding. OpenClaw does everything else.

simple nexus
#

It’s a fluid area. What is a harness? Do people really need ALL of OpenClaw if they have highly specific workflow needs?

sharp idol
#

Including coding

frozen zealot
sharp idol
#

I'm starting to think he's a paid promotor of Claude Code. lol

crystal jolt
crystal jolt
#

Like you guys aren’t helpful (I understand it’s not your job) but don’t tell me my questions don’t warrant anything of value?

lost gulch
#

Claude code is beautiful Openclaw is, well.. it’s just Openclaw.

marble ledge
#

@crystal jolt If you like ANT's models and their software that's great.

But Openclaw is an Architecture that is model/provider agnostic, and to certain extent, it's even harness agnostic. So you can use different models, multiple models, and even use some models in a different harness. And do it all at the same time.

It's open source, so you can hack on it, improve it, make it worse kek , and expand beyond what is already there and what it can do.

If it doesn't interest you or you don't find it useful, well, that's okay too.

simple nexus
#

They have different purposes. I don’t think Claude code is designed to talk to you over whatsapp, or have access to personal data like email ?

frozen zealot
boreal rapids
#

I've not used Claude Code, I'm sure it's very flexible these days. Does Claude Code let you use whatever models nowadays?

crystal jolt
crystal jolt
#

In my case I didn’t know about openclaw so I already built everything

#

Maybe if I knew about openclaw before then it would’ve been a different story

crystal jolt
craggy bane
crystal jolt
#

Well the latest of each model type, (haiku, sonnet and Opus) you usually have no reason to what the earlier versions of a model type since with every new version there is actual improvement unlike open ai slop

crystal jolt
boreal rapids
raven sundial
raven sundial
frozen zealot
raven sundial
#

for openai you need to use a proxy layer, idk about chinese labs except minimax they offer an anthropic capable endpoint

crystal jolt
frozen zealot
boreal rapids
#

I think in the end the harness is just a harness, with work you can extend any harness to do anything any other harness can do.

crystal jolt
#

are there any guidelines for models running on openclaw that the models don’t have when standalone?

solemn bone
#

i think v2026.5.7 is broken. message delivery always failed

raven sundial
frozen zealot
raven sundial
#

except it being slow while using a browser lol

solemn bone
sullen locust
#

yeah do thats

loud ravine
#

Gemini is fucking awful, it hallucinates so badly especially on the free version, that I couldnt stand it.

And whats worse, is when you point it out thats its wrong or hallucinating, it apologies and tries to correct itself by…posting the exact same hallucinated shit

frozen zealot
loud ravine
#

Though I too am noticing claude is starting to get bad with the hallucinations especially on niche subjects

#

They really need to train AI to say “I dont know” rather than creating fictional stuff

#

Claude was good at that, now its not

frozen zealot
loud ravine
#

I noticed with agentic AI, hallucinations drop down to “it either rarely does, or never does”

frozen zealot
#

specifically pre-training data

loud ravine
#

Or maybe have the AI actually read what it wrote and determine if its actually true or just bullshit

frozen zealot
#

but Gemini is different, the reason it hallucinates a lot is because of the sliding window attention in the local layers

loud ravine
#

I had gemini make up fake commands, or command arguments that doesn’t exist, because I was working with openclaw + gpt-codex-5.5, and I wanted a second opinion because I was skeptical

#

I pointed out to gemini these were not real

“I apologize, try this”

gives the exact same shit

frozen zealot
loud ravine
#

Gemini has the intelligence of the robot vacuum in Idiocracy

#

Where it just stuck and repeats itself while being wrong

#

“Your floor your floor is now clean”
“Your floor your floor is now clean”

tepid kestrel
loud ravine
#

I feel like AI is actually going to eat itself alive, it first peaks with training data that is good, it gets better with more training, then now with so much AI slop on the internet, the training data is tainted, and then in like a self fulfilling prophecy, the AI just cannibalizes itself

#

That would be ironic that AI ends up destroying itself

tepid kestrel
boreal rapids
#

Humans hallucinate just as much, just differently 😛

tepid kestrel
#

no our excuse are we call it "assumption"

raven sundial
#

or forcing creative and smart people into generating data to train itself?

tranquil mural
crisp nimbus
# loud ravine Gemini has the intelligence of the robot vacuum in Idiocracy

Tbh I haven’t really had this experience with Gemini. For me it’s been very good and quite impressive. Also it is in fact the only AI that has ever straight up told me “I don’t know”. Granted, my agent has a very defined prompt that also explicitly tells it to not assume things and use “I don’t know” when it truly doesn’t know, but even with this prompting, OpenAI and Anthropic models have never told me “I don’t know” and have made shit up instead…

#

Even the flash models surprised me for more general tasks they are very good (and fast). Heck even Gemma 4 has been used a lot by me and I don’t have many complaints although it clearly is no good for stuff that needs more reliability like more complex coding

#

Gemma 4 is more than capable of managing a Linux server in real time and not breaking things in an unrecoverable fashion (it has corrected errors it has made before)

#

Gemini in the official Gemini app is pretty shit though. Like, they keep breaking stuff and making it worse there I genuinely have no idea wtf they are doing to make it so bad

severe fulcrum
#

Gemini for coding has always lacking… they might stick with selling TPU 🫣

boreal rapids
#

I managed to get Gemma 4 26B A4B running on my GTX 1070 at a respectable 25 tok/s 😄

tepid kestrel
#

thats actually alot

frozen zealot
boreal rapids
tepid kestrel
#

what you do is, you use sonnet 4.6 for the planning with tasks and let gemini 3 flash to the job

frozen zealot
#

no, never trust gremlin 3 flash

boreal rapids
#

Go with goblin mini instead? 😄

severe fulcrum
crisp nimbus
severe fulcrum
boreal rapids
#

I also had Qwen 3.6 35B A3B running at 17 tok/s, it is quite a bit slower than Gemma 4, but I haven't compared the output results of more difficult coding tasks yet.

severe fulcrum
frozen zealot
crisp nimbus
boreal rapids
frozen zealot
zenith violet
#

After the recent openai token based usage update my limits started going down way too fast after reaching above 100k context level. Does anyone have a solution for this other than reducing the context length and force trigger compactions? The context mostly goes up that much due to tool inputs/outputs. been thinking maybe its a good idea to shrink them or summarize them to keep the context low

crisp nimbus
severe fulcrum
crisp nimbus
frozen zealot
crisp nimbus
zenith violet
crisp nimbus
zenith violet
frozen zealot
crisp nimbus
#

My weekly limit is gone and I’m on the plus plan lol

crisp nimbus
#

But I guess they all care about their high numbers on benchmarks rather than actually using the thing for themselves to test it

zenith violet
severe fulcrum
#

I personally use Stitch quite a lot for website designs it’s crazy good

crisp nimbus
frozen zealot
zenith violet
crisp nimbus
crisp nimbus
sweet zodiac
#

Locally AI is the app to use for local AI models on iOS/iPadOS

crisp nimbus
#

But my iPhone SE 2020 is too ancient to run anything above 1b decently.

frozen zealot
#

I have talked about it so many time, search your app store for edge gallery

#

it's available right meow

zenith violet
crisp nimbus
#

I’m still holding on on iOS 17.0 coz of TrollStore. It’s just so nice to not be fully in the apple jail while still enjoying the benefits of the walled garden

#

Also tbh I don’t like liquid glass… like at all. To the point that I even considered switching to android, but like, google is also locking down android so… at this point whats even the appeal?

dire tartan
#

I'm still on Iphone SE

#

and I love it

craggy bane
#

n 🧊

tepid kestrel
arctic trench
dire tartan
arctic trench
#

Those screens are tiny

dire tartan
#

I don't need a big screen to phone

arctic trench
dire tartan
arctic trench
dire tartan
arctic trench
tepid kestrel
languid scroll
#

Any ETA on the macOS companion app download? The beta button on the website links to GitHub source only

lean tangle
#

Can anyone playing with the recent beta's check if the bug that prevented OC from sending .zip files in discord DM's is fixed (please)?

steady hearth
#

Hello,
Need routing: where should I file a /voice bug report?

lean tangle
steady hearth
marble ledge
gilded obsidian
#

How does anyone use openclaw with a local model? the context of all the stuff it sends with each request is ... crazy... I'm sure there's a way to cache it model side somewhere (or maybe it already does)

lean tangle
gilded obsidian
#

I've got a nice phat model loaded up that is able to do 128k context even (Google Quanting it)

lean tangle
#

if you use something like openai's API, you can check the logs to see all the stuff OC sends

gilded obsidian
#

Hmm I've got it working now its just slower than I thought.. but hey I managed to Quant qwen3.6-35b-a3b-local onto a 16GB Nvidia 4060ti, and I can even fit the most important Experts from the MoE all on the GPU!

#

Takes almost a minute for response, I'm sure I can tweak some settings somewhere maybe get it faster, without any context I'm getting almost 40 tokens/sec tho 😄

glossy salmon
glossy salmon
honest current
ebon atlas
honest current
cursive widget
#

I need help in configuring and installing my open claw.

cursive widget
cursive widget
cold jacinth
cold jacinth
cursive widget
cold jacinth
honest current
cursive widget
#

No. I'm currently using Gemini.

cold jacinth
#

You have any coding harness?

#

On your personal computer

cursive widget
cold jacinth
#

What’s the model it’s running

cursive widget
cold jacinth
#

Ok can you ssh into that computer?

cursive widget
cold jacinth
#

Gonna make this a thread so we don’t spam the chat

cold jacinth
cursive widget
cold jacinth
short granite
#

I haven't updated in two weeks. Is it safe to update or is it broken?

cursive widget
#

Please try to explain to me As if I was 14 age and I'm not that smart.

cold jacinth
tepid kestrel
cold jacinth
short granite
#

Hermes Agent took number one slot on Openrouter probably because of all this update drama.

short granite
tepid kestrel
cold jacinth
short granite
#

wdym?

glossy salmon
cold jacinth
# short granite wdym?

Depends on your last install, but many plugins moved out of core package etc to keep Openclaw light and smooth.
Many of the plugins have dependencies that may or may not be updated. If the plugin fails, it can cause issues with claw

honest current
honest current
glossy salmon
cold jacinth
glossy salmon
#

wait even if I don't know who that is, you're listening to someone from Nebraska how to say "lasagna"? why not listen to an Italian since it's an Italian dish?

tepid kestrel
cold jacinth
#

It’s Spanish

glossy salmon
cold jacinth
#

Sorry it’s Greek

cursive widget
#

Which one is better, open claw or Hermes?

cold jacinth
#

Check it out

tepid kestrel
glossy salmon
cold jacinth
cold jacinth
cursive widget
glossy salmon
cursive widget
glossy salmon
cold jacinth
cursive widget
glossy salmon
# cursive widget Thank you for your feedback.

I quite literally asked for help in this server and I was told to use AI to fix a simple Oauth issue. Ai (Claude Code) tried for 50< minutes to fix it and then SSH tunnel broke so I gave up

cursive widget
tepid kestrel
cold jacinth
tepid kestrel
#

my advice

cursive widget
cold jacinth
#

Hermes more controlled as they only allow their own plugins so there is less to break. Openclaw more powerful and more customizable.

glossy salmon
tepid kestrel
cold jacinth
cursive widget
tepid kestrel
cold jacinth
#

@glossy salmon you probably think pasta is Italian

#

It’s Chinese

#

Basil? Not native to Italy

#

Roman Empire brought in a lot of things

glossy salmon
cold jacinth
#

Tomatoes? Not native to Italy

cursive widget
cold jacinth
cursive widget
glossy salmon
# tepid kestrel try antigravity

I did, i even paid for the pro plan, it sucks. the 5 hour limit is a lie and servers are always too busy to fill my request. Ive never been happier to switch to claude

cold jacinth
#

Turn that frown upside down

tepid kestrel
glossy salmon
tepid kestrel
#

due to some google issues, i couldnt even go far this week anyway

Screenshot_2026-05-10_014241.jpg
cursive widget
#

Helping with setup

glossy salmon
cursive widget
#

Please come to chat soon can see you screenshare.

cold jacinth
cursive widget
cold jacinth
tepid kestrel
#

you cursed me @glossy salmon

glossy salmon
tepid kestrel
glossy salmon
cold jacinth
#

My favorite part of Italy is New Jersey

cursive widget
#

@cold jacinth stay on topic.

cold jacinth
cursive widget
#

Are you trying to be irritating on purpose?

cursive widget
fiery nacelle
#

Isn't the topic "general"?

cold jacinth
gilded obsidian
#

Ugh still having issues with running a local model 🙁 It ran like 10-12 tool uses, crafted most of the code for "snake game in rust" then just... stopped no idea why, frickin qwen model is annoying me but its such a good model, I think it's just a tooling or not talking right back to the claw issue 🙁

#

It certainly got past context issue, I have the window at 32k and it would have bombed if that was the issue, the llama.cpp instance didn't crash either

fiery nacelle
#

What's this news about Goblins and GPT 5.5 that im hearing about? Lmao 🤣

marble ledge
fiery nacelle
marble ledge
fiery nacelle
marble ledge
ebon horizon
#

Do we know if theres plans to bring /goal into openclaw?

marble ledge
cedar prawn
#

My agent is "embodied" as a digital crow with goblin energy; maybe it's catching.

ebon horizon
#

Background task failed: ACP background task (run 15028540). Internal error

#

Anyone know the straight forward fix? Codex session spawning

fiery nacelle
#

Ha, was working on some custom changes in my OpenClaw, told my agent "alright, let's run some tests" expecting the usual "what shall we test today?" Response i always got. Nope, not this time. This time the response was "Let's run THE test." and went off and did something else entirely. Technically test passed, tool calls executed properly, didnt loop, proper responses, suspiciously extremely ambitious. 🤣

steel vector
fiery nacelle
marble ledge
fiery nacelle
# marble ledge Still bretty gud for qwen

Honestly one of my favorite local models so far. Enough for me to have it as my main model for like 90% of the things I use it for. Heavy coding just goes over to Codex.

marble ledge
# marble ledge Still bretty gud for qwen

I was using an older qwen3-coder-next months back when I first got started. first model I ever tried with openclaw. It wasn't as on point as the premiere models, but it still worked well enough to get stuff done. Especially with better prompting. Helped me learn stuff about how models work.

obsidian mural
obsidian mural
#

@fiery nacelle do you have suggestions of which codex models to use in paperclip ? eventually I will move to a local model but just want to get everthing working first

current i have gpt-4o and cheap model as gpt-4o-mini

fiery nacelle
obsidian mural
obsidian mural
polar hill
#

Hello

#

Hiring people who are good with posting tiktoks with Agents

gilded obsidian
#

@fiery nacelle Anything special you did with Qwen 3.6 to get it to work? I can run the 35B @ 128k context but man its hit or miss if it works, and gets hung up on tool calls after awhile

#

qwen3.6-35b-a3b (6-bit) on 4060ti & CPU via TurboQuant runs great from the cli at least...

#

I also noticed my MEMORY.md is getting huge, I would have assumed there is a way to tell certain workspace memories to only apply to that workspace automagically, but I guess not. Have full qmd running as well.

glad lintel
#

Did the fix the discord issue yet?

gilded obsidian
#

The one where it doesn't respond sometimes because it's trying to use a channel name instead of id or someshit?

marble ledge
gilded obsidian
#

I mean... the agent? Not sure what exactly you are asking? 😄

#

<!-- openclaw-memory-promotion:memory:memory/2026-05-02.md:1:10 -->

marble ledge
gilded obsidian
#

@marble ledge Really large code repos..

tender pawn
#

How much of a downgrade is going from sonnet 4.6 to got 5.5 instant

gilded obsidian
#

If it would just keep these memories in their prespective "channels" (discord) workspaces instead of all in here it wouldn't be so bad.

#

Keep in mind, I iz total newb

marble ledge
gilded obsidian
#

(to open claw that is, very technical person though)

#

@marble ledge Rules from config somewhere or simple system agent rules?

marble ledge
gilded obsidian
lean tangle
gilded obsidian
#

It hasn't mattered until now since the model I use allows for 1 million tokens, but now I'm starting to use a local model with 128~256k context....

#

@lean tangle Yes but let me check that it's actually done anything with it

lean tangle
marble ledge
lean tangle
#

There are knobs for tweaking the dreaming. I’m not sure what they are, but they’re in the docs.

calm quarry
#

Can computer use literally just use a computer like a human? Like it can use like Twitter without API????

gilded obsidian
#

hmmm why does Dreaming say it's doing all 3 stages of dreaming at the same exact time? Could that be an issue?

lean tangle
lean tangle
gilded obsidian
#

Dude these dreams are weird AF... "A crab with dollar signs for eyes sat at a coastal trading desk made of driftwood, whispering psssst — it's at four-twenty as the tide came in. The numbers floated above the waves like bioluminescence: 13.50 dissolving into sea foam, 402.50 carved into a barnacle-crusted rock, 420 glowing in phosphorescent green. I was canceling limit orders written on kelp, the ink running but somehow holding. A crustacean CPA with glasses too small for its eyestalks warned me about short-term capital gains while tap-dancing on a breakeven line drawn in wet sand. I asked how much I'd owe and the crab just clicked its claws — about five hundred more than you thought, stonk wizard. The pre-market bell was actually a buoy clanging somewhere past the fog bank, and I woke up laughing, still half-convinced my portfolio was denominated in sand dollars."

marble ledge
#

I need to dig into the dreaming more, tbh. I've been distracted by other things. Need to educate meself.

fiery nacelle
gilded obsidian
#

I mean I do use it for stock stuff too LOL

lean tangle
gilded obsidian
#

You should absolutely check out TurboQuant, you can fit a huge model these days 😄

#

I mean should I smack on backfill maybe? See if that clears things up?

calm quarry
# lean tangle Why not?

I am wondering does it still need to execute an API call? Or whether it can literally just use the computer/twitter like I would

lean tangle
#

This talk about dreaming has me tired. lol.

gilded obsidian
#

I'm not sure how to verify it is working, I do see lots of the items that are in MEMORY.md in the "Daily Log Review" section of dreaming...

lean tangle
#

Read the docs on dreaming….

gilded obsidian
#

oh lord that might put me to sleep, maybe I can tell my agent to summarize em 😉

#

So I'm having a crazy issue though where I've defined the Qwen3.6-35B-A3B as local, with a certain context but "openclaw gateway models" lists it as an openai one, with a completely different context, like its not even reading my numbers :/

fiery nacelle
gilded obsidian
#

DeepSeek is 75% off til the 31st.. when that expires I'm screwed unless I can get qwen 35b chugging

# 
          "alias": "Qwen3.6 Local"
        } ```  ```       "local": {
        "baseUrl": "http://localhost:8081/v1",
        "apiKey": "llamacpp-local",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen3.6-35b-a3b",
            "name": "Qwen3.6-35B-A3B",
            "reasoning": false,
            "input": [ "text" ],
            "contextWindow": 32768,
            "maxTokens": 8192,
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            }
          }
        ]
      } ```  ``` openai/qwen3.6-35b-a3b                     text       195k        no    no    configured,alias:Qwen3.6 Local ```
#

How does that even make SENSE?!

#

why is it 195k, why is it not considered local?

#

I've got it set to 32k for now because of testing...

#

In fact I don't think ANY model in openclaw.json has 195k context, so no clue where it's getting that number.

#

Heck let me ask the agent :/ 😛

marble ledge
#

I mean if the promotion was as aggressive as it sounds for you, but I know you can tune it.

gilded obsidian
#

@marble ledge 90% of it would be gone out of MEMORY.md if there was a simple setting somewhre (and there just might be and I'm an idiot) that tells it to save memories to the associated workspace they were created in if that memory seems like it's for that space and not a global one.

#

Ho boy he's writing an essay on the 195k context and where its coming from

marble ledge
#

So we have like the daily log with appended daily notes.

#

And then I use losless-claw for context management.

#

You said you already have qmd enabled, which I think it great. It's worked well for me anyway.

gilded obsidian
#

I already ahve memory rules all over the place..

marble ledge
#

Okay, well then it sounds like you are squared away. haha

#

Memory has been one of those things that everyone is very opionated on and has their vibe that works for them.

gilded obsidian
#

Sorry 😄

#

Wont let you post a bunch of MD.. I get it

marble ledge
gilded obsidian
#

I tried to post all my memory rules

#

server said no no, can't do that much markup 😄

marble ledge
#

Ahh okay. Do you find that what you have been doing as far as the rules have been working well for you?

ebon atlas
#

how tf did strickland just win that lmao

gilded obsidian
#

No but I think I'll try adding my workspace idea

#

I just need to word it perfectly.....

# 
Something's rewriting my config. Let me see the full current state.
OpenClaw rewrote my config — it normalized everything. Let me check why and fix it properly. ```  Oh boy that was it's response to what's overwriting the config
#

its actually not getting overwritten on disk, if anything is happening it's in process/memory

#

So something is marking it as openai instead of local, and 195k instead of my defined 32k... to what end.. hmmmmmm

sullen locust
#

guess who

sullen locust
#

mostly because of #

gilded obsidian
#

Yeah my bad 😄

sullen locust
#

you won't be punished

#

unless big lord Perry comes to beat us all

ionic wren
tepid kestrel
gilded obsidian
# 

195k context: I still don't have the exact source for this. The llama-server reports n_ctx_train: 262144 (256K), and the config was rewritten to contextWindow: 32768 (32K). Neither matches 195k. It's likely OpenClaw's internal default for OpenAI-compatible backends. I need to dig into whether there's an openai/models set or openclaw providers set command that lets me override this properly.```
#

Hmm so because I have it marked as openai-completions it just assumes all that???

sullen locust
gilded obsidian
#

I was under the understanding that just told it what "protocol" it was

polar hill
#

The Peter Steinberger has reacted to my message

tepid kestrel
polar hill
#

I feel honored

sullen locust
sullen locust
gilded obsidian
#

@fiery nacelle What do you have your local qwen set as?

#

@fiery nacelle Maybe we can help each other out here 😄

fiery nacelle
gilded obsidian
#

the "api" setting

# 
        "api": "openai-completions", ```
#

Apparently I have borked that, time to go RTFM (or copy off of you) 😄

fiery nacelle
# gilded obsidian the "api" setting 
      "lmstudio": {
        "baseUrl": "http://127.0.0.1:1234/v1",
        "apiKey": "LM_API_TOKEN",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen/qwen3.6-35b-a3b",
            "name": "Qwen3.6 35B A3B",
            "api": "openai-completions",
            "compat": {
              "requiresStringContent": true
            },
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 262144,
            "contextTokens": 200000,
            "maxTokens": 8192
          },```
gilded obsidian
#

ok so yours is the same...

#

except you have it there twice

#

I wonder if llama.cpp is delivering it differently than openai-completions understand

#

And how are you using reasoning = true, I thought that was a nono with qwen models

#

*pokes around llama.cpp's flags

fiery nacelle
sullen locust
gilded obsidian
#

Well maybe I'll just copy-pasta yours and modify the port number and provider name and see wtf blows up

sullen locust
gilded obsidian
#

models.providers.llama.cpp: Unrecognized key: "compat" Ah you are running ancient-claw aren't ya

sullen locust
#

maybe try uhhh

#

models.providers.openaiCompatible

gilded obsidian
#

or is there one specific to llama.cpp ?

sullen locust
#

then compact?