users-helping-users

[meager mortar]

i'm curious if people run ALL their AI prompts they might do through Clawdbot? I find myself thinking in three zones -- Claud app, Claude Code, and Clawdbot. I use telegram so far, so the UI and responsiveness of clawdbot is much less than Claud app. Still working through my use cases

[meager mortar]

my second question is whether anyone has setup different "personalities" of Clawdbot -- one tone in the morning, different at work, more mellow at night. I guess i could just explain that to one claudbot, but thinking whether i should have different Agents for differnt types of work or engagement i might be doing

[teal hearth]

I setup two personalities depending on who is messaging (me or my partner). So far it’s worked pretty well. It uses the sender’s identity to know “who it should be” when it responds

[low dragon]

/reset SOUL_EVIL.md xD

[velvet crane]

personally i use both clawdbot, codex and occasionally claude desktop app or chatgpt browser. its all context dependent. codex for coding, clawdbot for fun/assistant stuff

[meager mortar]

tip: i am putting my /commands from Claude Code in a place where Clawdbot can get them. I told Clawdbot not to store them in memory but to follow the instructions from the /commands. These are things like a daily checkin and triaging my todo list. My main reason for this is i want the same commands to be available in both f

[potent kraken]

Is the macOs companion meant to start the gateway? I currently just start the gateway from the CLI with "clawdbot gateway" but the companion app doesn't seem to latch onto it.

[light oak]

I have the same question

[meager mortar]

i had trouble with the MacOS from the DMG and it was last built 5 days ago, it hasnt been a priority. Is that what you are using? For me it seemed to connect to an already running gateway

[potent kraken]

it is what I am using but today in the process of fixing the anthropic auth I ran the gateway from the daemon and the macos app is guccis now. Except that it's crashing when opening the chat but I'm looking at that. Everything else seem to work

cf. #old-help message

[tough mason]

I put together a mac menu bar for managing the gateway and put it on github today, but don't want to just advertise stuff here. I didn't realize there was one already somewhere?

[mental mica]

yep! apps/mac, still wip while peter is putting out fires with the main process lol

(its a full app that also has a menu bar element)

#1458485243123208289

[tough mason]

Incredible what a LEGEND

[shell patrol]

I tried setting up on a vps with Ubuntu.
Telegram is connected.
Antigravity says rate limit is fine.

But aall i get is this error every 30 minutes

Cloud Code Assist API error (429): {
"error": {
"code": 429,
"message": "Resource has been exhausted (e.g. check quota).",
"status": "RESOURCE_EXHAUSTED"
}
}

[twin raptor]

i have been observing this annoying error again and again...my guess without knowing much is that your model name is incorrectly configured...i would run clawdbot models list to seee whats happenign

i pushed a vps/docker guide in the morning..have a look at that - https://docs.clawd.bot/platforms/hetzner

[sly niche]

same issue with antigravity, had to move to codex 😞

[shell patrol]

Oh thank you! Excellent, I'm also on Hetzner. Let me try that later.

I'll never get used to how discord is displaying threads 😅

[shell patrol]

Maybe the above guide posted by @twin raptor can help you too !

[analog gust]

yeah antigravity not working

[swift wigeon]

Hey y'all, anyone get Copilot sub working? just got this

[livid canopy]

btw hetzner has a docker image already, just install with that and you avoid extra steps

[leaden tinsel]

https://github.com/clawdbot/clawdbot-ansible

[livid canopy]

noice, i got mine running since a while but I was working on a one-click install for hetzner with pm2 and no docker but I stop cause this is good enough

[sour mist]

Codex through codex cli or droid? Any difference / which is better?

[old mortar]

codex cli works fine for me? never used droid personally

[serene trench]

Wow whatsapp is not easy to setup. 😅

[rocky orbit]

What is your issue?

[serene trench]

I finally got it. It didn’t work direct message. Had to create a group with my wife. Then remove her. Onboarding quick start path is broke for WhatsApp (I reported that on GH)

I happen to have WhatsApp but I don’t have two phones. Never thought that was the recommended way to see Clawd up

[serene trench]

So is the Mac OS app only accessible if you build from the GitHub repo?

[mental mica]

Yes

[serene trench]

Well then it's time to move to that.

[mental mica]

scripts/restart-mac is a good place to look for build stuff, I run that when I dev on it

Beware of dragons, it’s not fully finished yet. #1458485243123208289

[serene trench]

Yep, I'm okay with that !

Not really sure I want to keep my Mac mini from sleeping, and will migrate clawd to my Raspberry Pi since that's always on. However, while it's on my Mac mini, I might as well check out the companion app!

[mental mica]

It’s built so you can run the gateway on your pi and have the companion app on your Mac if you want!

Tailscale is King

[shadow igloo]

Tailscale is indeed king.

Since Clawd didn't know, do you guys have a hosted instance of the docker container anywhere? If not, I can spin one up, but it'd just simplify a bunch if y'all had it hosted in github or dockerhub or something

[mental mica]

https://github.com/clawdbot/nix-clawdbot and https://github.com/clawdbot/clawdbot-ansible both exist

[leaden tinsel]

https://github.com/clawdbot/clawdbot-ansible

[true tundra]

Having a weird issue - wanted to setup for replying without being @ - and it seems like it's kind of working but replying multiple times now ?

And if no @ he doesn't emote on it so I can't tell which thread it's replying to

Is there a proper config that I'm missing ?

[swift wigeon]

@silk ingot Do you have a guide for how to setup copilot? Is there a pairing method like other apps?

[mental mica]

Use #old-help with Krill

[swift wigeon]

Ah thx!

[serene trench]

Is it not advisable to use WhatsApp with the same phone number?

[mental mica]

no

bc it looks like you're texting yourself plus it gets all the messages you would get

[serene trench]

Yeah that’s not ideal.

[still crow]

whats wrong with that. + what if i want to get all of the personal messages

[mental mica]

thats your choice

[solar cedar]

Hi Guys, I need your insight.

I am running clawdbot on my Macbook sitting at home. I am connected to clawbot through slack.

It works perfectly. But after sometime, I guess around 20 mins. The slack message doesn't reply to my messages.

am assuming my Macbook went to sleep / got locked after a while. Anyone else faced this issue? How are you guys keeping the connection alive the whole day?

[honest barn]

any plan to have linux as companion as well to one central gateway

[brisk wadi]

Working on it. There is a github issue

[mental mica]

#1458489771797123225 too

[gritty zenith]

hey all! qq about memory, it seems like the daily memories are being created correctly, but the "long term memory" in MEMORY.MD has nothing in it (except one line at setup time)

is the line about long term memory in agents.md enough for ya'll? or should I add something to heartbeat or a cron job to make sure long term memory is updated?

[honest barn]

minimax m2

??

[mental mica]

I did a cron job myself

[gritty zenith]

makes sense! care to share what kind of prompt worked for you

did you include stuff about daily memories as welll

[honest barn]

might need to nuke the session or something i dunno.

[mental mica]

I just told mine to set one up every night to go through and save what it wanted to remember the most

[gritty zenith]

yeah maybe start a new session

[craggy sorrel]

I just setup Clawdbot and it was working out the box, but then I tried to change the models and I'm getting these errors. The string too long error is when I run clawdbot models set openai/gpt-5.2, and the 422 is when I switched to cerebras/zai-glm-4.7 (and same with 4.6). I couldn't find good documentation for getting this working but that's what I came across when I searched the Discord.

I do a lot of AI work but it's always possible I messed something up. This is my models configuration, any chance this is wrong?

"agents": {
  "defaults": {
    "model": {
      "primary": "cerebras/zai-glm-4.7"
    },
    "models": {
      "openai/gpt-5.2": {
        "alias": "gpt"
      },
      "cerebras/zai-glm-4.7": {}
    },
    "workspace": "/Users/mergesort/clawd"
  }
}

[crystal nova]

Hi guys, just a quick question. with the recent drama of Anthropic, if we use clawdbot with our claude models with max sub, will it be considered a violation of the TOS ? Are you guys using it after that ?

[pearl heath]

Can anyone confirm if the /clear command is actually real? I know that @sharp parcel used to mention it in the past, but when I run it the context stays the same.

I cloned the repo and Antigravity read through all the docs and it says that /clear isn't actually in the codebase and I should use /reset instead.

Any guidance to avoid hitting error 429 with anthropic?

[halcyon ether]

GLM 4.7 has issues with tokens; I ran into the same issue. I also tried to do clawdbot models after that and got a bunch of 422s; I think getting rid of model or models fixed it for me. Was a bit sleepy when working on it and didn't note it down at the time, sorry, hope this helps a bit

[raven mason]

I had this too, think it’s actually the context window getting too big. Delete all the session json files in .clawdbot and restart. You need to use /new to reset the context

[livid canopy]

hey guys despite having sessions configured per-sender everyhting goes into one big session.
Tried to debug with krill but no luck
ANy ideas?
nvm i needed some coffee

[tame pine]

Hey guys, just curious about your coding workflows, I find it a bit tough to code with Clawd in my discord channel as I'm used to cli in terminal with CC. However I feel like that way I'm missing much functionality that my PA has. How are you guys setting up coding workflows?

[compact oak]

Best way is to use subagents

Make a skill where clawd can read tmux panes

I can send my version if interested but it is pretty basic, also clawd wipes out the time problem with codex so I would almost always use it for coding as it is the goat at the moment

[tame pine]

I would definitely be interested! That's a good shout

[bleak vigil]

i tried sending clawdbot a screeenshot in telegram but it doesnt seem to read it and just responds with some hallucination or a screenshot from my mac. how do i fix this?

[vocal palm]

Now that it's been a day or two - does anyone have a rough idea how much a day of non-subscription use via Antigravity or Anthropic API costs? Rough ballpark?

[molten pond]

Tried to use #1459642797895319552 but the automated answers aren't helping
Since setting up a morning schedule for summarizing e-mail, my Telegram gets a heartbeat message every 30 minutes, non-stop
I'm not sure how to turn it off but it's super annoying 😂

[rocky orbit]

Tell Jarvis to stop it, he’ll stop!

[sand hornet]

Hey folks, is there a solution already for the "Claude Max subscription rate limited reached too fast" problem? It just burned through that with only a very lightweight 1 hour chat.

[potent kraken]

You could try to ask your Clawd to look over the days session and identify what is causing the most token consumption. You might have something you could optimise the token usage for.

I did that and turns out mine wasn't being very efficient with web search and looking up massive aria trees when it couldn't use summarize on some website and wasn't using qmd for memory / session searches

Made it a lot better for me

[tame tapir]

Is there a way to add both a dedicated Clawdbot What's App Number AND my private one? I.e., I would like for clawd to be able see all my conversations and while also being able to act under its onw number?

[serene trench]

I'm confused on how the bundled skills work. I have GOG installed, and Cloudbot doesn't seem able to access it.

I’m not seeing the GOG skill files in my workspace yet. If you know where they are on disk, tell me the path and I’ll load them. Otherwise, if you can share the download/repo for the GOG skill, I’ll place it at ~/clawd/skills/gog/ and use it to read your calendar

I can run gog manually in my terminal

So I know it's working and logged in.

[barren blade]

I think it depends on the model you’re using or the conversation tool (telegram, discord, etc). That has been my experience at least. I see GOG sometimes in onboarding and other times not. I maybe completely in left field, but that’s what I’ve seen

[serene trench]

Definitely using gpt-5.2 from my pro sub and telegram

I was never able to run the normal onboarding because I started with WhatsApp and the quick start onboarding kept switching to telegram. Then I moved to telegram because when I got WhatsApp working it was less than ideal. 😂

[stuck sparrow]

did /status break in whatsapp recently? I upraded/restarted gateway and /status doesn't do anything (works in telegram, not whatsapp), /thinking for example replies but not /status

[vivid grail]

I've setup and configured Claude Code, but still see in /models

[anthropic] auth: anthropic:default missing (+1)
• anthropic/claude-opus-4-5 (opus)
• anthropic/claude-sonnet-4-5 (sonnet)

how can I fix this?

[molten pond]

Anyone has any ideas about this endless heartbeat message im getting from Clawd to my Telegram?
Every 30 minutes 🙁 Been drivign me nuts and even when i ask to turn it on it says it did, but it keeps sending them

[meager mortar]

What’s App does formatting better. Telegram can have channels. I might try discord next

[serene trench]

Yeah but I don’t have two numbers and talking to myself is terrible. I think it feeds its own messages back to itself too.

Which has the best formatting? I think I can do iMessage too since it’s on a Mac.

[orchid gate]

anyone running their families with this? I need... hierarchy - for example I want to have it ping my daughter for a pic to prove her room is tidy every day and need loyalty (to me 🙂

[tulip dagger]

Legit got in my first ever fight with an AI tool - I authenticated Clawdbot with 5.2 Codex through ChatGPT OAuth. My understanding is that this means that my usage is through my Pro plan and not additional API usage payments. Claude, agreed with me. ChatGPT, agreed with me. Clawdbot was ADAMANT that I could still be incurring API costs without ever using an API Key and through my config. Finally got it to agree with me when I walked it step by step through the config, but out of caution, anyone else using 5.2 Codex through their ChatGPT plan can confirm? Usage is through subscription capacity?

[craggy sorrel]

That did it, thank you! I deleted them manually, what is the /new command? I'm ngl I think this project is super impressive, but definitely on the bleeding edge so I'm not always sure where to find things or notice the docs are incomplete. (Which I totally expect given the wild west nature of things.)

[meager mortar]

Same, I did WhatsApp for one day but didn’t like it (same number chat)

[raven mason]

If you just send your bot "/new" it will reset the context window 😄 save you deleting the session yourself

[craggy sorrel]

Ahh, thanks!

[serene trench]

How can you incur API costs if you don’t give your API key?

[serene trench]

The problem is she can tell Ai to “make this room look clean”

[orchid gate]

🙂 luckily we still don't have an honesty problem. That was just an example... I'm looking for something similar to a "Family account" with content restrictions and permissions similar to what Apple/MSFT/Google have. Doesn't have to be as fancy and strict... Thanks!

[tulip dagger]

Exactly what I asked Clawdbot

But it was quite stern at me insisting that OAuth didn't mean it wasn't transacting usage outside of the subscription

I even did echo, no APIKey was present. So I think I am good, but I wanted to ask

[serene trench]

As far as I know for most services that have API keys you’d have to explicitly put that in an environment variable (for example). There are some other methods but I think those are mostly for teams/enterprise. anyway yes the login with ChatGPT, not OpenAI API key. They’re two different services.

[orchid fog]

Hey folks -- antigravity was working just fine until I updated to the latest clawdbot version. Now getting the error below. In #1459642797895319552 I don't see a fix since yesterday. Would appreciate any guiance:

Cloud Code Assist API error (400): {
"error": {
"code": 400,
"message": "{"type":"error","error":{"type":"invalid_request_error","message":"tools.10.custom.input_schema: JSON schema is invalid. It must match JSON Schema draft 2020-12 (https://json-schema.org/draft/2020-12). Learn more about tool use at https://docs.claude.com/en/docs/tool-use.\"},\"request_id\":\"req_vrtx_011CX2fgs1gpvkaszwMogrdz\"}",
"status": "INVALID_ARGUMENT"
}
}

[fervent pivot]

Hey folks. New here. Just hit my codex rate limit. Does the Gemini cli set up use the Gemini pro membership?

[normal bough]

So I tried clawdbot with Claude code subscription and it works really well but it seems to be consuming lot of tokens. It consumed my max plan quota in 30min

[barren blade]

Turn it on or off?

[molten pond]

Will it not cause issues ?

[barren blade]

Mine pings me every 30 minutes to remind me discord (which it runs on) is taking too much memory. I need to change its settings but it won’t hurt anything

[solid sleet]

Hi, I was trying gemini-3-flash from antigravity and it couldn't do the tool call properly for setting a cron job. it was a simple reminder to ping me in 25 mins like a pomodomo session.
Got this error.

[molten pond]

The pings are fine I just don’t understand why these messages aren’t internal? Why is it sending them to me 🥺🤣

[molten pond]

I constantly get a dialog about node wanting admin permissions. Is there any way to grant things like Full Disk Access etc only to clawd, and not the entirety of node? Feels dangerous

[wicked jewel]

Does your HEARTBEAT.md file have any contents? You should be able to add some guidance in there to just return HEARTBEAT OK, which causes clawd to not message you for this.

Something like: "If nothing important to report to the user then return only HEARTBEAT_OK"

Or something more specific about what your heartbeat is doing here like "Do not bother the user about [whatever it is doing here], just return HEARTBEAT OK"

FYI, I'm still getting to know the system, someone else may have better advice

[wicked jewel]

Are people still using oauth rather than API for claude access, I switched to API after the drama a few days ago, but it gets expensive

[molten pond]

Personally switched to codex and minimax

[wary trellis]

Hey. Hope someone can help with this one.

I initially installed clawdbot on my Mac; ran perfectly.
Then wanted to run it 24/7, so installed on my Jetson (like a Raspberry PI, but different).

Used the new(?) "just works" installer (curl ... | bash), which gave a few errors (mostly permissions) as node/npm wasn't installed yet. Fixed that manually.

Configured clawd just fine.

But I keep getting

[clawdbot] ⚠️ Agent failed before reply: No credentials found for profile "anthropic:claude-cli".. Check gateway logs for details.

However, when I run clawdbot configure and go to model/auth it shows that the token is still valid for many hrs.

This GH issue seemed related, but didn't fix the issue

Thanks!

[wary trellis]

Pasting a token from claude setup-token did solve it btw

[sly island]

the fix for me was in ~/.clawdbot/clawdbot.json to replace "token" with "oauth"

[meager mortar]

you can get a year long oauth, i am on that for now

[meager mortar]

Its a bug that they are leaking into Telegram, there may be a fix already, idk

i posted this in another channel yesterday... clawd without Opus is pretty flat, but i can easily devour tokens! Am gonna use Clawd in pretty tight range -- morning routines where i want to be on my phone.

I feel at a crossroads with Clawd, I now have a mix of Clawd, Claude code, Claude chat app, and Gemini. Someone mentioned friction and that is it, mental overhead and with Clawd now I’m constantly worried about tokens. I love Clawd on Opus but almost feel guilty using that for morning check ins.

Writing my Oura ring scores to a markdown in obsidian is cool but worth it? (In tokens). Do i need Clawd to put on music or connect to app XYZ?

I’m worried about getting too hooked (good) and then Anthropic does a rug pull and/or limiting Opus even more.

[drifting helm]

Anyone has any ideas about this endless

[tulip dagger]

This just happened to me this morning while running a simple morning agenda cron job. Booooooo

[zealous raptor]

Not sure if this is related but hitting max usage quicky seems to be happening across the board https://www.reddit.com/r/ClaudeCode/comments/1qaqn0k/for_those_unaware_anthropic_is_kind_of_breaking/

[crisp wing]

r/ClaudeCode via Reddit

[lime gulch]

does anyone have a good agent skill or rules for react native development?

[drifting helm]

Anyone have trouble disabling heartbeats? I think I have them disabled, but I can't get them to stop (I have restarted the gateway)

[wicked jewel]

Was not aware of minimax, now just saw Peter mention it on twitter (https://x.com/steipete/status/2010761076598849759) - which model is best? I see minimax-M2.1 (+ lightning) and minimax M2. First 2 seem to emphases coding whilst minimax-m2 mentions agentic capabilities / function calling / reasoning etc?

[crisp wing]

@steipete via Twitter

Peter Steinberger (@steipete)

Been running Clawd on @MiniMax_AI the last few days after optimizing the implementation and it's a really great alternative.
︀︀
︀︀Now recommending this over Anthropic. For coding subagents you can still pick e.g. codex as subagent. MiniMax is really close for 5% of the costs.

**💬 41 🔁 7 ❤️ 244 👁️ 8.9K **

FxTwitter · Jan 12, 2026, 5:09 PM

[warm thorn]

Hey guys - How do you set Clawd up to be able to edit config etc locally, but also have it hosted remotely like on a VPS? Do you keep them in sync via github or something?

I'm running it on my laptop but want to be able to access it online when my laptop is off too

Thanks!

[random mauve]

git works great! That's what I used for my vps setup. https://docs.clawd.bot/concepts/agent-workspace#git-backup-recommended-private

[wicked jewel]

Trying to install latest version onto a linux box, ran into this error when onbaording after selecting minimax

run error: Error: Unknown model: minimax/MiniMax-M2.1

Any ideas?

[rocky orbit]

Did you try to get an answer on #1459642797895319552 ?

[wicked jewel]

Nope. But I got claude code to take a look and it says:
" This is a bug in clawdbot's setup wizard. The models.json file is missing the required apiKey field:

• Has: baseUrl, api, and models array
• Missing: apiKey (required by the model registry validator)

The setup wizard saved your API key to auth-profiles.json but forgot to reference it in models.json. The model registry validation fails silently, so your custom MiniMax model never gets loaded.

Fix

I need to add the apiKey field to your models.json."

Adding here in case it helps anyone, this allowed me to progress with onboarding

[harsh spade]

gm guys, sorry if this question has been asked already but my head is exploding and i cant figure out how to solve this issue lol. so i've installed clawdbot on my vps, connected telegram and it responds to simple queries. i've asked it to list the available skills and it says i have 22/51 ready. When i ask on telegram however a question that involves one of the skills, it says it has difficulties accessing the "weather" tool, to name one skill. how do I solve this issue?

[mental mica]

try asking krill in #1459642797895319552

[harsh spade]

ok thanks

[warm thorn]

great, thanks! Really helpful to know I'm headed in the right direction!

My plan is:

• Run Clawdbot on the VPS all the time (Docker)
• Use Git to sync the workspace (~/clawd) between laptop ↔ VPS.
• Use Tailscale for access
• Let Clawdbot edit files on the VPS, then pull those changes to my laptop when I'm back.

[undone dawn]

Right now when you do a / command in Slack like /status you have to first tag your bot, and then it seems like it actually sends the req through an AI and sends it back. This means formatting is always different, or Clawdbot will decide to leave some info out, or itll condense it etc. Anyone know how to fix this?

[clear granite]

Hi, I just discovered Clawdbot today and I'm trying to install on an AWS ubuntu server. When I ran this command
cd clawdbot && pnpm install && pnpm run build
I got a bunch of typescript errors:

Is this normal? Should I ignore it and run onboard?

[void jetty]

did anyone got nano banana working without api? i have a google ai pro sub

[long parcel]

this worked for me as well, thanks! ❤️

[dense glen]

I don't know where to talk about this so I'm gonna try here. I'm not sure if what I'm doing is something I should be doing. It isn't working out too great.
I want coding agents. For large, long running things. That I control over Discord or whatever. I've got it to spawn two extra agents that I bind to specific channels and all that jazz. But after that it just kinda goes meh. I tell it to do things (reminding it off the tmux and coding agent skill) and it does something. I see in tmux it's done. Nothing happens. It waits. It doesn't seem to make smart decisions most of the time either.

Krill told me to write in the agents.md that it always has to poll the tmux. It did for a bit. I wonder if I expect too much right now/from LLMs in general? It seems like it's almost a good idea but not quite

[orchid slate]

Regarding new memory thing: any help on this?

"One important gotcha: upstream memory search needs an OpenAI API key (OPENAI_API_KEY or
models.providers.openai.apiKey). You have OpenAI Codex OAuth, but that does not satisfy embeddings. Do you already
have an OpenAI API key set on this machine? If not, we’ll need you to add it yourself (I won’t paste secrets)."

[low dragon]

I am looking for a maintainer for this. I don't use Antigravity/Google stuff and it constantly breaks

[low dragon]

orchestration is hard and still not well-tested. That's something that needs experimentation.

[low dragon]

What is the question here? Message pretty much explains it?

[dense glen]

Figured. Well, I'm gonna experiment I guess! hah

[low dragon]

does this help? I asked codex to explain it

[low dragon]

that is weird. How did I not see any when I tested the install?

[low dragon]

fall back to git as simplest fix

[low dragon]

no that's a bug and pls try main I fixed a bunch on that today

[low dragon]

I fixed this on main alreadyt

[clear granite]

I am using the hackable version with pnpm. I spent hours trying to get to work with slack. I finished the onboarding process but I'm getting this error:

[mental mica]

#1459642797895319552

[sullen chasm]

I just read on X that MiniMax is supported now (youhou!), compare to Opus 4.5 is it worth it? how much it costs with a day to day use?

[wicked jewel]

I recently switched to minimax 2.1 based on this and am personaly having quality issues with it, interested to hear how others are doing that have switched.

It hallucinated it being evening even though the system clock is 11:40 am here.
Told me gog calender auth had expired and I needed to renew. I got claude code to look over all logs, that turned out to be false as well
Also asked it to write 5 paragraphs about something, but would only write 4 over 2 attempts.

Perhaps it works well for some things, but I think I need to switch to something else. Anyone having issues?

(This is via direct API)

Which unfortunately had a minimum initial credit load of $25

[verbal minnow]

Hi All,

Has anyone tried using qwen-3 coder

[olive laurel]

What is easiest way to install clawdbot on non tech person mac?

I was assisting to a few people but they don’t even have node or brew
Any pointers?

Many non-tech folks are interested to use it but need a lot of assistance
Happy to own. Need some assistance

[rocky loom]

I'm going a bit crazy with all the changes... Now suddenly my sub agents are running with different providers (I never set it up like that) - how do you all stay sane? How can I just not let al changes in config happen?

[wicked jewel]

Yeah the release cadence is dizzying! I just hold off until theres something 'must have' in the release notes. Definitely cant keep up with it at the moment

[rocky orbit]

The best would be to check #freshbits quite often and updates only if there is a critical bug corrected or a new features that you want to have.

[rocky orbit]

The doc from here could help you : https://docs.clawd.bot/start/getting-started

[rocky loom]

Yeah ill probably do that but sometimes something looks useful in an update and 5 other things break

[olive laurel]

So for tech people its easy to figure out
I was referring to non-tech people
Like business people or creators etc

[rocky orbit]

Tell them to come here and our friend Krill will help them on #1459642797895319552 for sure!

[bleak vigil]

is it just me or the latest update just broke a whole bunch of stuff? all my long running tasks no longer works. all the background agents are not getting triggered. etc. what happened? @sharp parcel

[rocky orbit]

Krill does not have access to this channel, only on #1459642797895319552 🙂

[idle jacinth]

has anyone gotten clawdbot on an ubuntu machine to work with

sandbox stuff

sandbox + browser

[elfin aspen]

clawdbot uninstall bash: clawdbot: command not found
funny how the uninstall does not work

[random mauve]

Yeah both work great. Did it on a hetzner box running Ubuntu.

[idle jacinth]

got it

i tried it out earlier and had some issues with the sandbox setup

i was trying to dockerise everything

gna try it again now that i know it'll work

[random mauve]

Tbh I just had a Claude Code instance do it all for me. But https://docs.clawd.bot/gateway/sandboxing docs are pretty fleshed out on this topic now too.

[idle jacinth]

thank you kind sir

🙏🏻

[idle jacinth]

does the setup_token thing still work?

for anthropic models?

[elfin aspen]

so if the uninstall wont work i need to hunt the files by hand ?

[bronze lion]

good morning folks! i have 2 questions see if anyone can help.

1. i run clawdbot out of my mac mini on my network but i mostly work on a different computer, can i access the gateway / config / menu bar app from my main computer and not from my mac mini?

2. is there a documented way to create a "second" agent? for instance if i want to give it access to different tools/phone number etc. i want one for my wife and one for me. any advice?

[rocky orbit]

Did you try to ask Krill on #1459642797895319552 ?

[granite galleon]

im getting "LLM request rejected: messages.5.content.0.thinking.signature: Field required" when trying to use minimax m2.1 through openrouter, did anybody see such an error before?

im getting weird bugs all over the place with clawdbot

weird, it disappeared after removing antigravity models even though i'm using openrouter.

[real gull]

FYI - @low dragon when installing clawdbot from npm, getting error with post install script.

⏺ Bash(rm -rf ~/.clawdbot && npm uninstall -g clawdbot && npm install -g clawdbot@latest)
⎿  Error: Exit code 1
npm warn deprecated npmlog@6.0.2: This package is no longer supported.
npm warn deprecated are-we-there-yet@3.0.1: This package is no longer supported.
npm warn deprecated gauge@4.0.4: This package is no longer supported.
npm warn deprecated node-domexception@1.0.0: Use your platform's native DOMException instead
npm error code 1
npm error path /opt/homebrew/lib/node_modules/clawdbot
npm error command failed
npm error command sh -c node scripts/postinstall.js
npm error Error: context mismatch at line 400: expected " }", found " }"

[real gull]

Also looks like it's reported here: https://github.com/clawdbot/clawdbot/issues/861

[real gull]

If anyone runs into it, looks like there's a PR: https://github.com/clawdbot/clawdbot/pull/871

[bronze lion]

i didn't let me try that

good call

[barren blade]

I’d go ask in #1459642797895319552 it depends on how you installed

[barren blade]

Probably better to go to #1459642797895319552

[meager mortar]

@low dragon with the new threaded #1459642797895319552 maybe this channel is “end user chat” or the like, more comparing notes not help. Just a thought. Good luck on your keynote!🍀

[elfin aspen]

i used the "curl -fsSL https://clawd.bot/install.sh | bash"

[gray chasm]

Anyone facing a multiple message spam in discord? I cannot figure it out for the life of me. Clawdboy basically always duplicates it's answers to me in my server.

I'll ask it something. It'll send me the eyes emoji. Respond. I'll type a follow up but in between I'll get spammed by the same answer again as if it processed my previous message multiple times

[sudden solar]

I've noticed there is a bird skill that interacts with X

Is there something similar for interaction with other social media? Specifically Linkedin, Facebook and Instagram

[full dirge]

was anyone able to make amazon work with clawd?

[warm thorn]

🤔 Any idea why this section of my clawdbot.json returns 'invalid config'? I'm adding it via the web UI:

{
  "wizard": {
    "lastRunAt": "2026-01-13T14:09:46.472Z",
    "lastRunVersion": "2026.1.11-4",
    "lastRunCommand": "doctor",
    "lastRunMode": "local"
  },
  "auth": {
    "profiles": {
      "anthropic:default": {
        "provider": "anthropic",
        "mode": "token"
      }
    }
  },
  "agents": {
    "defaults": {
      "workspace": "/home/ethan/clawd",
      "heartbeat": {
        "every": "0m"
      },
      "model": {
        "primary": "anthropic/claude-opus-4-5",
        "fallbacks": [
          "minimax/MiniMax-M2.1"
        ]
      },
      "models": {
        "anthropic/claude-opus-4-5": {
          "alias": "opus"
        },
        "minimax/MiniMax-M2.1": {
          "alias": "minimax"
        }
      }
    }
  },
  "models": {
    "providers": {
      "minimax": {
        "baseUrl": "https://api.minimax.io/anthropic",
        "apiKey": "${MINIMAX_API_KEY}",
        "api": "anthropic-messages",
        "models": [
          {
            "id": "MiniMax-M2.1",
            "cost": {
              "input": 20,
              "output": 60,
              "cacheRead": 1,
              "cacheWrite": 4
            },
            "contextWindow": 1000000
          }
        ]
      }
    }
  },```

[noble sorrel]

clawdbot-newbie here, I managed to instuct main agent howto fetch electricity prices (Finland), also howto command my Tesla (set scheduled charging based on cheapest hrs), I saved all this information (told agent to save) into TOOLS.md, is this right approach?

[potent kraken]

If that this the whole file I think it’s because you have a trailing comma at the bottom for “models” closing bracket and you are missing the main final closing bracket

[warm thorn]

thanks! that was just the portion I changed, that was causing the issue. I've resolved it now:

#1460960560148648048 message

[signal tartan]

"clawdbot daemon install" starts 2 instances.
MacOS as VM in UTM
Version: 2026.1.13
WrapperApp to give node FDA in SystemSettings

Tryed this:
Cleanup hint: launchctl bootout gui/$UID/com.clawdbot.gateway
Cleanup hint: rm ~/Library/LaunchAgents/com.clawdbot.gateway.plist

AnyOne how run in this issue?

Or is this normal, because i've to agents on different workspaces running?

[ivory rapids]

fox

[fleet acorn]

I have installed clawdbot and i also installed ollama but it keep getting a trim error
before the update the bot and llama worked but the bot wouldnt reply. nopw it replies but the trim error is the reponse

how do i fix this?

[rocky orbit]

Did you try to ask on #1459642797895319552 ? 👌

[steady hornet]

Hey Community, I´m getting this weird json-style responses and it also won´t perform any actions.
I did a complete reset and also a fresh install.
No change in behavior.

[azure hull]

I made the potentially disastrous decision to try to re-install clawdbot (while saving some artifacts like skills and memories) because I was having a ton of issues.

I'm trying to re-install and now I'm unable to use my claude pro account. I don't want to switch to the API because I love the tool and I know I'd go broke with pay as you go with how much I was using clawdbot. Has anyone else run into this issue? Any advice on if OAuth is still a valid path with Anthropic?

Additonal edit
I asked Krill and it said:

The quick fix: Use an API key instead of the CLI token method for now. Go to console.anthropic.com, create an API key, and configure it```

[rocky orbit]

@azure hull anthropic token Claude CLI is OAuth with Claude

In the clawdbot config

[bronze lion]

krill is down. it's the end! 🙂

[bronze lion]

ok maybe i'll ask here since krill is down and i've spent an hour trying to figure it out. i keep getting

Agent failed before reply: No API key found for provider "anthropic".. Check gateway logs for details.

when chatting with my agent, despite having picked Anthropic CLI and having re-done the authentication through the onboarding flow

anyone knows why? if it's relevant the most annoying part is this just randomly happened, started in the middle of a conversation with my agent where things were going fine

[rocky orbit]

Have you checked the config in your clawdbot.json in .clawdbot ?

[woven blaze]

this is known issue, update your clawdbot.json config to use mode oauth instead of token

i beleive mode needs to be oauth for claude code token or you get that error

[bronze lion]

oh ok, any idea why it would have randomly changed? let me try that now. thanks

thanks that worked! sorry that is so simple but i couldn't figure it out i kept re authing within the wizard

[rocky orbit]

No worries mate! That’s why we have this community

[gaunt pagoda]

I'm experiencing two related issues with Telegram channel:

1. Tool calls/results leaking to chat
   When I make tool calls, the raw JSON (both the call and the result) is being sent to Telegram instead of being hidden. The user sees things like:

[Tool Call: gateway (ID: xxx)]
Arguments: { "action": "config.get" }

2. Message flow seems to block
   After a tool call, subsequent messages don't seem to be delivered. It feels like only one message gets sent and the rest of the response is lost. The conversation "stops" mid-thought.

Environment:

• Model: google-antigravity/claude-opus-4-5-thinking (thinkingLevel: low)
• Channel: Telegram
• Clawdbot version: latest (npm installed)
Expected: Tool calls should be internal only, and all response parts should be delivered sequentially to Telegram.

Any ideas?

[spring hinge]

Can bird be used with wsl ?

[neat iron]

are you using antigravity as a provider? it's a known problem with google models. See https://github.com/clawdbot/clawdbot/pull/905
and https://github.com/clawdbot/clawdbot/pull/894

[gaunt pagoda]

Thank you for the reply! Do you think this happens on Claude 4.5 Opus through Antigravity as well?

[neat iron]

i use opus through antigravity and it never happens for me, it's really a gemini problem

[gaunt pagoda]

Can it be some config on my end?

[hot belfry]

is it possible to have 2 instances of a clawdbot, a prime one spoken to via DM will full access to everything on my personal computer, and a second one with limited access to certain folders for group convos with friends? all under one clawdbot instance and the same connections for Signal/Discord/Whatsapp?

(I could obviously create a second agent I guess running on a VPS but I don't have 2 x Signal accounts / spare cell numbers currently)

[hot belfry]

oh i just use the sandboxed part on a VPS right? somehow clawd didn't tell me that existed my bad

[meager mortar]

I asked clawd to do a top to bottom review of my setup. Found a couple of “big” problems like a duplicate Memory.md file and a bunch of tweaks.

So incredible to have a self correcting system!

[vale breach]

sorry I can't seem to find the answer to the question and it's probably been answered to death

did clawdbot get whitelisted to use anthropic subscriptions? If we're still waiting for an answer, where should I follow to keep my eyes peeled?

[mental mica]

No whitelisting has happened

[low dragon]

https://docs.clawd.bot/concepts/multi-agent
https://docs.clawd.bot/concepts/channel-routing
https://docs.clawd.bot/gateway/sandboxing
https://docs.clawd.bot/multi-agent-sandbox-tools
https://docs.clawd.bot/concepts/group-messages

[low dragon]

its TS so should be

[delicate blaze]

I'm getting lots of failed tool calls with opus 4.5 through antigravity 🙁

[wicked ember]

How do i get the clawdbot mac app to be a node for the gateway running on a VPS?

[naive geyser]

Anyone’s telegram mess up recently. Haven’t updated but noticed it’s like the topic ids got mixed up or refreshed

[inner bison]

Hi Folks. Clawdbot was working well for me until now. I get the error message "429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed your account's rate limit. Please try again later."},"request_id":"req_011CX8riL3xanFxveFEozQpE"}"

What should I do?

[normal sapphire]

hey guys just wanted to ask - does minimax work with telegram? encountering issues right now and the commands i got in the #1459642797895319552 thread from krill weren't really working

[warm thorn]

Has anyone managed to get minimax working well? Seems to be really dumb - I can’t tell if the thinking isn’t set right

[digital ingot]

set up a fallback model

[dapper mauve]

Hi! I’m trying to figure out the best way to allows clawd to access sentry. I have few options like mcp, somehow browsing stuff, creating or using a cli.. is there any overall consensus on what’s best? (for instance I’d avoid using mcp if possible)

thanks a lot! clawd is incredible!

[misty light]

anyone have a working "flow" to have clawdbot type messages to codex sessions and summarize responses (and watch for when codex is finished and notify me)?

even better would be able to manage multiple sessions + notifications

[boreal spear]

Heya - 0ldN00b here.

If i was to use clawd to cleanup email, set calendars etc ... so cool ... but i just cannot get my head around the privacy from letting openAI suck it all up. Can local LLMs be powerful enough to handle this yet?! Or do I need 10 mac studios and a nuclear power station to actually do this at home?

[inner bison]

So folks here, what is the default model you use and the fall back model? Do share please so that the rest of us can learn too.

[zealous raptor]

Maybe in a few years

[ruby plinth]

Has anyone managed to get the message flow working better in telegram?
i.e when you ask the bot to do something, it can acknowledge / ask any questions before it does an action?
right now seems like the acknowledgement and the action/action results/summary all kinda come in at same time (after the agent has done the action) - i tried via prompting / system instructions but always seems to send all the response messages together at the end (hope that makes sense?)

[mental mica]

there's a block streaming mode config, i dont remember it exactly but ask your clawdbot

[ruby plinth]

ah ok, that might help - i see what you mean - might give that a go - just wanted to avoid where we get lots of messages all the time (like only stop to ask on first message /request - was worried if i enable the block streaming then all messages will come in - like "ping, ping, ping" even when just longer responses - also: block stream might not allow it to pause to ask questions etc - instead just show the stream in different blocks rather than all together - was more kinda logic than message legnth if you know what i mean?

[mental mica]

ah yea

i dont know the specifics of that part of the code but you can always /stop manually

[ruby plinth]

like if i prompt claude code to ask clarifying questions before taking actions then i ask it to do something it will not carry out the action unless its clear or there are not any questions... that sort of thing

[potent kraken]

For this I think you need to change the queue mode on top of enabling block streaming:
https://docs.clawd.bot/concepts/streaming#block-streaming-channel-messages
https://docs.clawd.bot/concepts/queue#queue-modes-per-channel

not 100% on whether block streaming is needed but you definitely need to change the queue mode

[ruby plinth]

ah great thanks 🙂

[lapis laurel]

hi, i'm having issue with claudecode auth. i've even resetted mac mini for this. i get FailoverError: No API key found for provider "anthropic". Auth store: /Users/gankim/.clawdbot/agents/main/agent/auth-profiles.json (agentDir: /Users/gankim/.clawdbot/agents/main/agent). Configure auth for this agent (clawdbot agents add <id...
any known solution?

[faint wagon]

Anyone have a script for ensuring that clawdbot gateway is always up? Sometimes clawd "kills himself" and this is especially annoying when I'm on an airplane 😉

[faint wagon]

https://discord.com/channels/1456350064065904867/1461441134302859315

I forgot how smart 🦞is

[limber rapids]

Anyone know if there’s a way to turn this on? My agents can’t seem to see each others messages on telegram

Or is it just not possible on telegram ?

[real gull]

What do y'all use for Voice Transcription? Been trying to use a custom skill that sends it to Gemini 3 Flash but it constantly tells me it's not working after working for like 2-3 hours lol

(I use Signal, mostly - or the TUI)

[mental mica]

Turn off privacy mode on your bot in botfather

It can only see replies and mentions with it on

[limber rapids]

It’s off

[limber rapids]

It’s a telegram public policy

No bot messaging

[mental mica]

Oh, I thought you meant you couldn’t see other users sorry I didn’t read that fully

[limber rapids]

I’ve tried a ton .. session send doesn’t work cos they on different gateways

Tried redis pubsub ❌

Finally got agent to agent working using webhooks

[sinful dawn]

Have any of you guys seen the user not found 401 error from open router? The account is funded and the key is fresh but not brand new. Definitely valid.

[undone edge]

Anyone getting this today after updating?

clawdbot@bot:~/instance/clawdbot$ pnpm build

> clawdbot@2026.1.15 build /home/clawdbot/instance/clawdbot
> tsc -p tsconfig.json && tsx scripts/canvas-a2ui-copy.ts

src/auto-reply/reply/get-reply-run.ts:178:13 - error TS2552: Cannot find name 'saveSessionStore'. Did you mean 'sessionStore'?

178       await saveSessionStore(storePath, sessionStore);
                ~~~~~~~~~~~~~~~~

  src/auto-reply/reply/get-reply-run.ts:120:5
    120     sessionStore,
            ~~~~~~~~~~~~
    'sessionStore' is declared here.

src/commands/models/auth.ts:283:33 - error TS2304: Cannot find name 'ClawdbotConfig'.

283 function applyDefaultModel(cfg: ClawdbotConfig, model: string): ClawdbotConfig {
                                    ~~~~~~~~~~~~~~

src/commands/models/auth.ts:283:65 - error TS2304: Cannot find name 'ClawdbotConfig'.

283 function applyDefaultModel(cfg: ClawdbotConfig, model: string): ClawdbotConfig {
                                                                    ~~~~~~~~~~~~~~

src/gateway/server/__tests__/test-utils.ts:4:9 - error TS2741: Property 'providers' is missing in type '{ plugins: never[]; tools: never[]; channels: never[]; gatewayHandlers: {}; httpHandlers: never[]; cliRegistrars: never[]; services: never[]; diagnostics: never[]; }' but required in type 'PluginRegistry'.

4   const base: PluginRegistry = {
          ~~~~

  src/plugins/registry.ts:89:3
    89   providers: PluginProviderRegistration[];
         ~~~~~~~~~
    'providers' is declared here.

src/plugins/providers.ts:8:54 - error TS2339: Property 'config' does not exist on type 'PluginLoadOptions | undefined'.

8   config?: Parameters<typeof loadClawdbotPlugins>[0]["config"];
                                                       ~~~~~~~~


Found 5 errors in 4 files.

Errors  Files
     1  src/auto-reply/reply/get-reply-run.ts:178
     2  src/commands/models/auth.ts:283
     1  src/gateway/server/__tests__/test-utils.ts:4
     1  src/plugins/providers.ts:8
 ELIFECYCLE  Command failed with exit code 1.

[low dragon]

that should be fixed

[low dragon]

I'm working on that

[low dragon]

If you wanna be unblocked you can always spin up a local agent and type "fix build" or wait a few min

[clear granite]

How can I prevent Clawdbot froming sending me messages that don't mean anything? I tried multiple times to tell not to do that and even gave it a communication window.

[steep vale]

when I sent /stop, it goes on until completing the current task and reply as if /stop is the next msg?

[shell bison]

Clawdbot Tool Schema Bug – Google/Gemini/Antigravity Models

Error Message

[agent/embedded] google tool schema snapshot
[agent/embedded] google tool schema has unsupported keywords
[agent/embedded] google tool schema has unsupported keywords

Issue

Tool calls fail when using Google/Gemini/Antigravity models. The error appears twice (likely for browser and canvas tools). Other providers (OpenAI, Anthropic) work fine.

Suspected Cause

The tool schema (JSON Schema for tool parameters) contains keywords not supported by the Google/Gemini API. Possible candidates:

• Tuple-style items
• additionalProperties
• $ref or complex nesting
• Certain type combinations

What We've Tried

• Checked out commit b7ba94f0c ("harden antigravity claude support" #968) – build hangs in TSC
• Checked out origin/main with commit 3dddbe105 ("fix: ignore properties in google tool schema warnings") – build also hangs
• Related PRs: #926 (tuple items), #741 (unsigned tool calls)

Current Workaround

Switch to Anthropic or OpenAI:
bash
clawdbot models set anthropic/claude-opus-4-5

Question

Has anyone found a fix for this? Are there specific schema keywords that need to be stripped for Google/Gemini compatibility? Any guidance would be appreciated!

[quiet igloo]

hey everyone. just getting started . I just did the /start and didn’t get the code to pair my telegram bot to my vps. Not sure what i am doing wrong. Any help is appreciated

[neat iron]

i couldn't get the pair message either. I just added my telegram ID to the allowFrom list; you can ask @idbot on telegram to get your id if you don't know it

"telegram": {
      "capabilities": [
        "inlineButtons"
      ],
      "enabled": true,
      "dmPolicy": "pairing",
      "groupPolicy": "allowlist",
      "streamMode": "off",
      "reactionLevel": "extensive",
      "accounts": {
        "main": {
          "dmPolicy": "allowlist",
          "botToken": "8232is",
          "allowFrom": [
            6793
          ],
          "groupPolicy": "allowlist",
          "streamMode": "partial"
        },
}

did minimax toolcalls stop working for anyone else? it's like antigravity models now 😭 just plain text

[versed adder]

I'm getting crushed with rate limiting

Think I overloaded my bot w/ too much info and loaded up the memory

[rigid charm]

Greetings people. Anyone else had issues with regards to browser lag?

the browser interface in Chrome keeps crashing chrome. Have to quit chrome and restart - but if I scroll through my chat a bit, gg. And Safari is so lagged up that I can't even type now. Literally after just a few interactions.

[noble sorrel]

Hi! Has anyone configured gog cli on a Raspberry Pi successfully? I have tried to do the oauth by tunneling SSH to the raspperry and opening the oauth callback URL on my mac, it works but then on on the gog cli side it just says "Fininishing..." ie. gets stuck.
I also tried to export tokens from mac gog cli to the raspberry then gog auth tokens import but it gets stuck too.

[noble sorrel]

Solved! keyring need to be set to file gog auth keyring file

[sudden solar]

ive been having issues on the initial setup. I have my clawdbot on VPS and it using the GPT4.0 model for now. However any task i give it. It says its doing and then just never completes. I try to install a skill from UI and it doesnt complete. I feel like the whole install is missing some persmissions or I went off track along the way. Even a simple cron job request doesnt happen.

[zenith solar]

have you checked the logs?

[muted cipher]

Anyone on the latest pull from github who is getting this error on opus antigravity

[1/16/26, 1:14:13 PM] Jarvis: ⚠️ Agent failed before reply: Request was aborted. Check gateway logs for details.
[1/16/26, 1:14:13 PM] Jarvis: Message ordering conflict - please try again. If this persists, use /new to start a fresh session.

[sudden solar]

yeah they just lead down rabbit holes and i feel like trying to patch something that shouldnt be broken to start with..im prettty sure im doing something wrong on the VPS

[zenith solar]

Does anyone have the android app working?

I've built it from source and installed it on my phone but can't get it to connect to the bridge (see https://discordapp.com/channels/1456350064065904867/1461670902994571347)

[mental mica]

It’s very wip still, #1458485323553181900

[versed adder]

What is everyone doing to prevent burning through their tokens in 10 minutes? If I talk to the clawdbot And go back and forth for a few minutes. Ask it about a feature or something. I know that it is using tokens to read the code and to logic. But before my session this morning I cleared the memory, or compacted it I should say, and it ran out in 10-15 minutes. I have it on automatic through the Claude CLI, I know that's not ideal. Is the way to handle this is to configure a cheaper model to chat with. And have claudebot just call other models to do other work? I know I'm missing something simple. I'd appreciate any feedback. I have the $20 subscriptions for Google openai and Claude, my plan is to burn through my weekly limit, which I'm sure I'll do by the end of the day today or tomorrow and then upgrade to the max plan. But even with that, I would. I would toast whatever lemons they gave me on the $100 plan especially.

Sorry I'm in my car using voice to text. I promise I know how to write in English.

[winter folio]

Any suggestions or recommendations for backing up ClaudBot’s memory/config/etc. I had a small issue setting up slack today and it would have been nice to have an easy rollback.

Git for the .clawdbot directory seems like a good first step, but with keys/tokens I wasn’t sure about pushing it to GitHub (private repo)

[mental mica]

i do github private repo for the workspace folder but i dont backup my sessions or config right now

[timid knot]

It seems the gateway dashboard chat ui (dubbed "agent:main:main") is a different harness/agent config from the one connected over telegram? I otherwise can't make sense of why it is reporting different things

[dense glen]

I'm getting a lot of that today. Not quite sure why.

[muted cipher]

for me its completely broken, it doesnt even reply and its stuck forever 😆

[dense glen]

Yeah a /new fixes that but it happens again soon after

[muted cipher]

I was using that for sometime, until it also stopped working, also weird, I reverted back to yesterday's commit and nuked all the session, still its completely broken, no messages and nothing is logs

edit: had to do a full reinstall to fix this and it works now

[honest linden]

has anyone been able to format inline buttons for telegram?

[mental mica]

wdym format inline buttons

[honest linden]

i mean so the bot is able to send messages with inline buttons

tried a few times but doesnt seem to work

[mental mica]

it should work, have your bot look at the codebase and troubleshoot itself

[fiery shoal]

hey all 👋 ☺️
quick question: how are you currently, or how would you advise, building and maintaining a data storage for the agent to use?

"use" means:

• daily adding things like bookmarks, github stars, meeting transcripts, documents, etc
• embedded vectors for storage and retrieval (got 20k+ documents)
• ideally self-hosted

i'm using Supabase but that's more because of legacy, prototypes built over the last 3 years; i'm thinking Convex or sqlite; maybe obsidian?

what do you all use?

[mental mica]

i just use the built in clawdbot stuff with the md files and have it sync that to github

what are you storing

[feral fjord]

@silk ingot is there a way to share specific discord channels within this server with my personal clawd bot (read only) so it can keep up with and summarize all the actiivity for me

huh, ok clawd bot not in here i guess

[fiery shoal]

@mental mica , some of the things:

• twitter likes and bookmarks
• github projects which I star
• work stuff like meeting transcripts and summaries, Notion pages
• pdfs

Basically, as much context as I can, centralised

[mental mica]

#server-faq

[feral fjord]

ive been lookin in there 😂

[mental mica]

its the Can you add my bot to this server to keep up with updates? entry

[feral fjord]

ah ha! thank you!

[knotty pasture]

Anyone know why clawdbot sometimes responses include spam messages in telegram?

[mental mica]

that looks like it accidentally linked a domain

[knotty pasture]

you're right! The response included README.md which links to that spam site

thanks

[celest iris]

With beautiful things like the clawbot chrome extension i begin to see already that clawbot should run not in a sandbox on my device. I did set it up originally sandboxed as part of the wizard, how can I switch it to run on the full system? I'm on Fedora 43 KDE Edition Linux

[limber cape]

anyone know how to get around the loop where on a new install it keeps looking for a claude api key instead of using my claude code max sub

[hot belfry]

best way to tweak prompts with specific guidance for Signal, Discord, other channels etc ?

[limber rapids]

Doing something similar

Do share what you land on
I had do a chromaDb in the past pre the new semantic dB integration by default

Each agent has an index of memory session skills

The I have a bigger dB with all my meetings and notes with all agent memory

[tulip dune]

Hello everyone. I have tried to get this to work probly 30 times now. i am getting this error from the gateway.
[canvas] host mounted at http://0.0.0.0:18789/__clawdbot__/canvas/⁠ (root /home/node/clawd/canvas)

[canvas] host listening on http://0.0.0.0:18793⁠ (root /home/node/clawd/canvas)

[bridge] listening on tcp://0.0.0.0:18790 (node)

[heartbeat] started

[gateway] agent model: qwen/qwen3-coder-30b

[gateway] listening on ws://0.0.0.0:18789 (PID 7)

[gateway] log file: /tmp/clawdbot/clawdbot-2026-01-17.log

[browser/server] Browser control listening on http://127.0.0.1:18791/⁠

[telegram] [default] starting provider (@SamanthaAIv1bot)

[telegram] setMyCommands failed: GrammyError: Call to 'setMyCommands' failed! (400: Bad Request: command description length must not exceed 256)

[ws] unauthorized conn=9b23a8fb-f682-4519-915c-4405d577c1b3 remote=172.19.0.1 client=clawdbot-control-ui webchat vdev

[ws] closed before connect conn=9b23a8fb-f682-4519-915c-4405d577c1b3 remote=172.19.0.1 fwd=n/a origin=http://localhost:18789 host=localhost:18789 ua=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36 code=4008 reason=connect failed

if anyone has any ideas please let me know.

[hollow sky]

I can't get anthropic auth back to be working. It just dropped yesterday and every time I try to auth it, I always get: ⚠️ Agent failed before reply: OAuth token refresh failed for anthropic: Failed to refresh OAuth token for anthropic. Please try again or re-authenticate.. Check gateway logs for details.

Gateway logs shows invalid auth token even though it was just newly generated and set.

Previously it worked fine. Anyone else had this issue? Krill couldn't help.

[hollow sky]

ok, I had to manually pull it from keychain and set it, onboard didn't pull it from keychain, dind't ask me for permissions

[celest iris]

Turned out that i fucked over my rate limit with using thinking budget high for cronjobs. The more you know.

for rather deterministic cronjobs, do you guys use a thinking budget at all or do you have thinking turned off?

[neat iron]

for deterministic crons i set it to use minimax (soo cheap)

[celest iris]

when you have a main model (GPT-5.2 in my case), you can have it use seperate models for crons simultaneously?

[neat iron]

yea when you create the cron you can override pretty much anything. I just set up an agent that only does crons

clawdbot cron add \
  --name "Deep analysis" \
  --cron "0 6 * * 1" \
  --tz "America/Los_Angeles" \
  --session isolated \
  --message "Weekly deep analysis of project progress." \
  --model "opus" \
  --thinking high \
  --deliver \
  --channel whatsapp \
  --to "[phone]"

more at https://docs.clawd.bot/automation/cron-jobs#cron-jobs

[modest night]

Guys

Is there a way to prevent bot from going silent after replying once? Example, bot says it will read the docs, but instead of replying, it stays silent until I message it again, or even several times, before replying. Sometimes it even forgets.

[neat iron]

check the session or logs; does it have a tool call or something? for me that usually happens because it failed to do a tool call and aborts

[modest night]

That's right, bot said it's waiting for the tool call to return the result. But sometimes the tool has finished its task, yet it remains silent, or even invents a response.

[fiery shoal]

For now, I have a self-hosted Supabase on Hostinger, and made two MCP tools to retrieve tweets and git repos, embed and store (https://github.com/duhman/tweet-vault and https://github.com/duhman/star-vault), plus a daily cron job to update; but this feels over engineered and there must be something easier. I'm gonna try next the following combo: Obsidian (easy to read as well) plus this skill https://clawdhub.com/steipete/markdown-converter

[celest iris]

Cheers

[hoary nimbus]

How do I turn off heartbeats entirely? I keep getting messages that heartbeats.md is empty.

[potent kraken]

you set the frequency (agents.defaults.heartbeat.every) to "0m" in the config. https://docs.clawd.bot/gateway/heartbeat#defaults

[hoary nimbus]

Thx can I do this from the control ui?

[elfin aspen]

i been playing around with this and i find APITA to use the local config, is there a better documentation how to setup the models for local use ? also swaping them, loading 200+ GB models takes a while

[neat iron]

you can go to config on the web UI, switch to raw and edit that value; then save and restart

[hoary nimbus]

Ty! I looked but must have missed the value. Good to know!

[neat iron]

if you have just one agent it should be under agents.defaults; if you have multiple then it can be under agents.list[];
you are looking for something like

        "heartbeat": {
          "every": "30m",
          "model": "minimax/MiniMax-M2.1",
          "target": "last",
          "prompt": "Read HEARTBEAT.md. Follow it strictly. Do not infer or repeat old tasks from prior chats."
        },

if it doesn't exist you can just add it and set to 0m! "heartbeat": { "every": "0m"}
more examples here: https://docs.clawd.bot/gateway/heartbeat#heartbeat

[tulip dune]

hey guys were can i locate the ui token aparently the token dont match and thats why my gateway isnt running. i can find the token in the clawdbot.json but i can not locate were the UI token is located so that i can match them.

[timid knot]

@neat iron interestingly my heartbeat (on the default setting) is quite chatty 😅

[mental mica]

do you have thinking turned off?

[timid knot]

I did turn off verbose just now to be sure but it actually just replied right after

@mental mica uhmm, /status shows thinking low, yes (:gear: Runtime: direct · Think: low)

[mental mica]

ah, if the model has thinking turned off, it'll find creative ways to still be able to think, e,g doing it in the regular message

[timid knot]

ah, what would you set it to then?

high?

btw I'm now curious what y'all have in your HEARTBEAT.md

[crisp wing]

The link provided is invalid, or the platform is currently not supported. :(

[tawny fox]

📥 *Download: ~9.4 Mbps* (25 MB in 21.2 seconds)

That's quite a bit slower than the first test - might be network congestion or Cloudflare throttling. Want me to run another test with a different server?
[19:12, 1/17/2026] Andrei Tanase: Speedtest again please I tested manually and get 500 mbps
[19:12, 1/17/2026] Clawd: [clawdbot] Just posted the results above! ⚡

*Download: ~9.4 Mbps*

That's a bit slow - could be:
- Network congestion
- Cloudflare throttling the test endpoint
- Or actual connection speed

Want me to try a different speed test method?``` Hmm anyone knows why this is happening?

[mental mica]

thats not a clawdbot issue lmao thats just whatever with your machine

[tawny fox]

FIgured it out, it was using Invoke instead of curl so I guess it was clawdbot, lmao

[scenic horizon]

Hey, has anyone experience managing clawbot across two instances (the plan was one on may daily driver laptop, onether on a stay at home proxmox omarchy vm)? Or should I just setup two of them?
Got this from the clawdyguy themself.

Practical options

• Single responder (recommended): Run Clawdbot on Omarchy only; when you say “on mac”, it SSHes into the Mac and runs commands. Mac can be asleep; Omarchy still answers.
• Two bots: Separate Telegram bots for Mac vs Omarchy; you choose by which bot you talk to. Cleanest routing.
• External fallback logic: Use systemd/health checks so the Mac instance only starts if it can’t reach the Omarchy instance. Still not perfect, but reduces double replies.
Update needed? Not for any of the above. Unless you want newer features, we can stay on current version.

[mental mica]

run the gateway in one spot, use the #1458485208721395843 (most are WIP) and ssh over tailscale to connect to the rest of the machines

one gateway, one memory, multiple control points

the recommended one that it had

personally i have an old imac that lives at home that mine runs on and tailscale to connect to everything else

[meager mortar]

i was hoping to get incremental (draft) typing of messages in telegram or discord, but couldnt get either to work. The technique is the bot continually edits and existing message until finally hitting send.

On discord i was closest, but it has rate limits that prohibit. On telegram, i could not get it to work. Perhaps it is a new telgram feature rolling out or something -- it says it needs " private chats with topics enabled (forum topic mode)" - which i cannot figure out what that is. I am giviing up for now, but if anyone gets it working, let me know!

[mental mica]

telegram should work, lemme cook

but yeah not easily possible on discord in a normal chat scenario

[meager mortar]

one thing that might be going on is that a Thinking model (Opus) takes time and makes a full answer, so then i was testing with Sonnet. But mostly i use Opus (for the quality) so it might be a moot point

[mental mica]

this is what i was thinking of

[manic gull]

Is there a consensus on the right base plan to run Clawd? I have $20 subscriptions for ChatGPT and Claude but my instinct is to create a new one for minimax just for clawd? Would the $20 plan be ok?

[mental mica]

I have the $20 minimax as my fallback

I never had any usage limits when I had to use it for two days

[meager mortar]

Right, and so that is not implemented in Clawdbot yet? Its not totally clear to me if it is fully implemented in Telegram. In addition: " sendMessageDraft is a new feature in Telegram's Bot API 9.3 (released December 31, 2025). This method allows bots to stream partial messages to users while the message is being generated. The bot must have forum topic mode enabled in private chats to use this feature. [repeating some of what you showed]

[mental mica]

that screenshot is from pi checking the clawdbot codebase

i havent tested it myself, im on a branch working on another platform

[tulip dune]

i am using windows 11 and docker desktop i have been trying to get this to run for the last 3 days and i cannot get past the 1008 unautherized problem i have tried docker install npm and pnpm allways the same issue. it has to do with the dameon i just cant get it to run. it says the token dosnt match. is there anyone that has a solution to this problem or can direct me to an answer.

[glad ibex]

Anyone have any trouble using gog to read emails and calendar? Weirdly, I was able to get it working when using claude opus 4.5 but I switched to gemini 3 flash (openrouter) and it's not able to access despite multiple auth attempts

[cosmic isle]

I was having the same problems, not sure if its because of tooling calling or some weird gemini quirks.

[jovial thistle]

Check your config under gateway.auth I personally just set it to disabled or open or whatever to avoid that 1008 error

[sacred swallow]

Hi guys,

I’m using clawd with whatsapp - I have given him my own number and chatting with myself on mobile web which it replies perfectly fine

But my notifications have stopped working completely- I do not get any from anyone

As soon as I stop the gateway, notification work fine

First I though it was the ack reaction setting causing this which I disabled but still no luck with notifications

Is it because I have given the number as mine?

Thanks

[topaz valve]

hello, I must be dumb, but I don't find how to architecture something that seems to be quite basic :

• my remote Linux Debian is hosting my Clawdbot (thank you so much for this amazing piece of crustacea... hu software)
• I want to link my MacBook when it's up and running, to let Clawd query my Mac-specific stuff (Things, Reminder, ...)
  Both are linked with a working Tailscale

Clawd itself suggests to configure my Mac as a node and pair it to my distant gateway, But I'm a little bit lost as "clawdbot node" command isn't working anymore, and the macOS UI app is said not to manage nodes yet...?

Another question : do I have to maintain two gateways on both sides and link them ? I never succeeded to do that, and don't exactly understand if it's useful for my case.

[mental mica]

the macos app is a node, but also just tell your clawdbot to ssh over tailscale

[fast ermine]

Currently working with local Minimax, anyone ever experienced the tool calling being parsed incorrectly?

E.g. “what’s the weather” gives:

minimax:tool_call
<invoke name="read">
<parameter name="path">/home/nas-pi-0/.npm-global/lib/node_modules/clawdbot/skills/weather/SKILL.md</parameter>
</invoke>
</minimax:tool_call>

the raw vLLM output returns it properly

(Notice first XLM row its <>)

[fair raptor]

I thought it was possible to oauth in via a Claude pro max sub in setup. Looks like anthropic has completely blocked this now ?

[mental mica]

Works for me on latest

[floral sand]

It's working for me, using it right now.

[fair raptor]

It’s just the setup if you’re in perhaps it works fine, I’m going through setup now

[barren dove]

hey guys. i have a subscription for claude code max and gpt pro. is there a way to use any of them when using clawd? when i try to use claude it says that my api key is not configured and when i try to use gpt i t says that my organization is not verified

[floral sand]

When I went through the install it asked me if I wanted to use my Claude subscription

As for the GPT error I saw that too, it's only for certain models I think.

[fair raptor]

You have to specifically prompt for setup token on oauth

[fast ermine]

If anyone gets this, use openai-completions in the config for response type

[flint mica]

Anyone have issues where you’ll be talking to your clawd and it keeps replying to a old message? This is a example

[tulip dune]

Thanks i will give that a try

[glad ibex]

So did you find a fix? Or switched back to Claude ?

[hot belfry]

best Gmail MCP or other integration to use with clawd?

[rocky orbit]

Gog cli I would say

[hot belfry]

why does claudecode keep pinging deadends in the docs? any tips to avoid this?

(there seems to be something inherent in the doc structure (is it the internal # links?) that claudecode struggles with and therefore makes it bad at debugging)

[mortal oyster]

Forgive the double post - also in "help" for Krill:

I'm tying myself in knots trying to work out why I'm furiously burning tokens, even when not using clawdbot. Have exhausted my daily Claude max, Codex, and some Openrouiter credits on top. Heartbeat set to every two hours, but Clawdbot digging around seems to think it is in some sort of carzy burn loop and is being called in the system prompt. Any suggestions welcome. Clawdbot on mac mini, node on MacOS, mainly using Telegram to communicate to clawdbot. in case helpful, here's one of the reported investigations:

ROOT CAUSE IDENTIFIED 💡

The "heartbeat prompt" is being automatically appended to system event messages (like "Exec completed", "Node connected", etc.)!

When the system sends notifications like:

• System: [timestamp] Exec completed (lucky-wi, code 0) ...
• System: [timestamp] Node: Tim's MacBook Pro ...
Those system messages are followed by the heartbeat prompt text. And each one triggers a full heartbeat run.

This means:

• Every time a background exec completes → heartbeat prompt → agent run
• Every time a node connects → heartbeat prompt → agent run
• Every time there's a system event → heartbeat prompt → agent run
That's why there were so many heartbeat runs!

Where is this coming from?

This is either:

1. Clawdbot's internal system - the gateway/session manager is automatically appending the heartbeat prompt to system events
2. A channel plugin/automation that's configured to send heartbeat checks after system events

---

Anyone got any tips?

[valid relic]

which models do you use ? i face many issues with claude API , and catching up errors on groq and other models

[celest iris]

Can minimax 2.1 be used with clawdbot via the coding plan or does it need to be API?

[rocky orbit]

Yes it can!

[celest iris]

Cheers, because I need to get into the weeds of configuring clawdbot in more details with using models outside of the main model. otherwise as a european regular joe the token costs will eat me alive 😢

[floral sand]

What models work best besides Claude? Claude works great but I keep using up my quota (even on the $200 plan). I've tried many other models but I haven't found one that works as well as Claude sonnet or opus. Quite frankly they all seem "too stupid" to understand and work within the complexity of the Clawdbot system. But Claude takes to it like second nature.

[celest iris]

GPT-5.2 (normal one) via codex cli's auth is amazing - but I too hit quota too easily, that's why I'm now getting into configuring it in more detail

[floral sand]

I find Gemini 3 Pro to be pretty decent in all other applications but with Clawdbot it's super stupid.

[celest iris]

my plan is to have skills, cron and heartbeat and eventually other tasks run via the cheaper model and keep 5.2 as the main model

[floral sand]

Claude sonnet and opus are the only two I've been able to get to work really well. GPT-5.2 I guess is again just too expensive and will run out.

[celest iris]

apparently it's more efficient than Opus with clawdbot, but like anthropic it will run out

[floral sand]

Maybe I'll give it a shot... what kind of plan do you need?

I was thinking that one of these cheaper reasoning models like R1 or Kimi K2 would work really well but R1 barely functions, K2 is better but not great.

[celest iris]

@floral sand I asked about minimax's coding plan, because the concensus in the discord seems like minimax runs best for "cheap" rather non-deterministic tasks like heartbeat, your cronjobs etc. So I'll give them to minimax. if it's bad, it was 20€ of play money and learning stuff about configuring AI agents, if it does work well, all the better

[warm thorn]

‪Is there a resource or notes somewhere on best practices to get the browser and logins working on a vps? I’m working through options with clawd but wanted to ask ‬

[floral sand]

Thanks I'm going to give it a try. $20 is definitely what I'm looking for.

[floral sand]

I just set up Clawdbot on a VPS, I can try to help. Can you elaborate on browser and logins? Do you mean getting the macOS desktop app to connect?

[tulip dune]

does anyone know how to fix this error? i have been banging my head for the last 3 days. help dosnt help been through the documents and repo plus a few AI helpers and i still can not figure this out. im having problems with the gateway keep getting 1008 the gateway wont run, clawdbot anything dosnt run. this is the error
[ws] unauthorized conn=ec6dd392-d827-4a55-a06b-9e1c71882f2b remote=172.19.0.1 client=clawdbot-control-ui webchat vdev

[ws] closed before connect conn=ec6dd392-d827-4a55-a06b-9e1c71882f2b remote=172.19.0.1 fwd=n/a origin=http://localhost:18789 host=localhost:18789 ua=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36 code=4008 reason=connect failed

[ws] unauthorized conn=4a3a9924-82cf-4095-9c6f-77dd639f1dc0 remote=172.19.0.1 client=clawdbot-control-ui webchat vdev

[ws] closed before connect conn=4a3a9924-82cf-4095-9c6f-77dd639f1dc0 remote=172.19.0.1 fwd=n/a origin=http://localhost:18789 host=localhost:18789 ua=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36 code=1008 reason=unauthorized

[bitter jetty]

I’ve been having Clawd do live testing on my app and it’s absolutely DEVOURING Anthropic credits, burns through my $200 plan in one run.

Anyone tried something similar with a different model that worked?

[old wharf]

“What have you tried yourself as in troubleshooting this so far?”

[red quail]

are you using bun? i did and whatsapp was buggy but swapped to node and now all fune

[old wharf]

@low dragon

[glossy sand]

Using GLM 4.7 via code plan. Works very well, although a bit slow.

[rocky orbit]

Opus burns a lot of token. We suggest using cheap models for simple tasks such as Haiku or Minimax

[glossy sand]

What is the best way to have Clawdbot “see” images?

[celest iris]

did you compare it to minimax? myself I now figured, instead of GPT + additional models, I start out minimax first, see how that goes and reintroduce GPT for things minimax sucks at.

[rocky orbit]

Did you try to work locally via Claude or Codex on your terminal?

[celest iris]

vibes wise minimax is more fun than GPT for sure, it's noticable how much they distilled claude loool

[glossy sand]

Haven’t tried MiniMax. Tried codex a bit, Gemini 2.5 flash via OpenRouter and now GLM

[frozen bolt]

Hey everyone - is there a definitive guide on how we can restrict access to secrets from agents? I keep getting them reading .env files and config files , logging/printing, etc and can't seem to make it stop?

[neat iron]

ask them to edit their core files making it non negotiable to access env files and to never print these things

[frozen bolt]

And rewrite on updates?

[rocky orbit]

You can stash and repush on updates from your bot

[native burrow]

Has anyone figured out how to use opus 4.5 through google vertex w clawdbot

[tulip dune]

yes i am installing it 100% local but I did set it up to use lmstudios (i dont have a claude or any sub) i do use open router from time to time but really trying to keep this local. im a complete novice so there is a very high probability that this is on me missing something.

[valid kettle]

I'm trying to login to vertex, do I need to use these environment variables (from g cloud), and where do I need to set them in the config?
https://github.com/badlogic/pi-mono/blob/ad337e30be87ecc8794468fd03fc264dfe23cfc9/packages/ai/src/stream.ts#L68-L71

[floral sand]

@celest iris I just tried minimax works really well. It’s fast and the quota for only $20 is very generous! Great solution

[celest iris]

I switched my plan up, now it's my driver and I will reintroduce the other models for other stuff, but I'm a bit skeptical. I do like it in english as well, but just right now as you finished this sentence I had my first true idiot moment when installing a skill.

so I'll look next into how/where/why to setup a more powerful model for skills to use, so minimax will call the skill with the different model or an agent to run the skill or however that works 😄

[floral sand]

I’ve been trying everything and nothing compares to Claude in my experience

[modest estuary]

I see many questions about token usage here. Even those with $200 subscriptions seem to burn through them quickly. This weekend, I asked some questions and set up some skills, but my tokens are gone as well. How feasible is it to use something like Clawdbot if tokens are consumed so easily? I was excited, but now I don't know anymore

It's frustrating that I have to consider these things, to be honest. Should I now carefully plan which models to use for each question?

[sly star]

Also feeling the token churn on haiku, ripping through my 5hr limit in around 15-20 mins while trying to squash bugs in clawdbot source code while leveraging the clawddocs skill. i reset the context as often as plausible and run sub agents with haiku to offload context bloat from the main session. Would love to get some real use out of clawdbot but its not looking promising with this level of token churn.

On another note, I'm runing clawdbot on an older 2019 intel Macbook Pro it seems like there are lots of bugs with the tui around tool card display, session switching footer updates, etc anyone else having problems with the tui?

[dire spade]

Yo, are yall coding with clawdbot directly or asking it to spawn coding agents in codex/claude code?

[floral sand]

Try using a cheaper model like Minimax. It’s not as good as Claude but it’s affordable at least.

[floral sand]

I’m using Claude code in tmux. I can access it and Clawdbot can access it.

[modest estuary]

I will try thanks

[dire spade]

Are directing clawdbot it to use cc in your terminal session? I’m just starting out, but clawd has only ever done coding tasks within its current session

[flint mica]

@small nest

It still hasn’t saved memory

[small nest]

Now i get only 1 answer.
BUT:
he only answer when i write. No auto-reply
and no memories or something.
Only old stuff.. like my name or something+

So.. did you got auto reply working?

[flint mica]

Wym by auto Roslyn

[small nest]

Or maybe any cronjob?

[flint mica]

Reply*.

[small nest]

for now, he only writes once when i write him first

like.. claude opus writes again without "ping"
example:

Me: Hey, can you check something?
Bot: Ofc, let me see....
Bot: yes i have checked it.

[flint mica]

Like this?

[small nest]

for me it works like this:
Me: can you check something?
Ofc, let me see....
Me: and?
Bot: yeah i have checked it. here is the answer

[flint mica]

Wait, my session-memory hook wasn’t enabled

[flint mica]

He does it sometimes

[small nest]

can you test something?
Write me in 10 seconds

What will he do?

For me he does nothing.. only when i write him again first

[flint mica]

Yeah it’s a weird bug. Sometimes mine does that sometimes doesn’t

Watch logs when it does it

[small nest]

But maybe something wrong with Heartbeats..

[small nest]

well.. at least its better than mine haha

whats your Heartbeat?

[flint mica]

I’m digging deeper into the memory issue rn

[small nest]

okay

one thing.. how did you create the background service for visual code?

my terminal is blocked right now because its running code tunnel

[flint mica]

I don’t even remember I made Claude do it for me while I was at the movie theatre 😂

[flint mica]

It’s blocked?

[small nest]

haha

[small nest]

yeah, because its running in the front

[flint mica]

What terminal is blocked. You sue copilot thru cli?

[small nest]

i tried to setup everything with perplexity gemini 3.0.. but its so lost

[small nest]

server terminal

[flint mica]

You gatta set it up as a background job

[small nest]

I m using clawd@silk ingotbot:~$ code tunnel --accept-server-license-terms --name clawdbot-tunnel which talks with a visual sudio code in my browser

because i m on a headless server

Whats your setup?

[flint mica]

quote from my bot:

I found the issue. You're right - it's the copilot-proxy causing it.

What's happening:

1. AGENTS.md instructs Larry to keep daily logs at memory/YYYY-MM-DD.md
2. Larry WAS creating these when using direct Claude API
3. Since switching to copilot/claude-haiku-4.5 proxy, Larry stopped following those instructions

Why copilot-proxy breaks it:
The copilot-proxy provides Claude via OpenAI-compatible API, which may have:

• Different sampling parameters (temperature, top_p)
• No extended thinking or prompt caching
• Less consistent instruction following for complex workspace behaviors

Your options:

1. Switch back to direct Claude API (recommended for full behavior):
   docker exec clawdbot-clawdbot-gateway-1 node dist/index.js config edit

Change agents.defaults.model.primary to "anthropic/claude-sonnet-4.5"

2. Use session-memory hook (already enabled, but only triggers on /new):

• The hook is working - it just needs you to run /new to save sessions
• Files will be 2026-01-18-slug.md format with LLM-generated slugs

3. Keep copilot-proxy but manually prompt memory updates:

• Periodically remind Larry: "update today's memory log"
• Add explicit reminders to HEARTBEAT.md

The copilot-proxy is great for cost savings, but sacrifices some of the nuanced behavior that makes Larry proactively maintain
memory logs. Want me to help switch back to direct Claude API, or adjust the setup to work with copilot-proxy?

[small nest]

well when i tried to save something manually, it does not save it

[sly star]

Same here, I get a lot of trailing off and have to constantly poke for a response

[flint mica]

ima switch back to my cladue code and see what happens

[small nest]

okax

sad if thats really for nothing... useless stuff :/

[small nest]

maybe we can use a script or cron to "sync" the memories

[flint mica]

Won’t solve the issue of it repeating itself tho smh

[small nest]

well, i dont have the repeat problem, so there should be a fix for it

Maybe you need to activate "selfChatMode": true,?

[18.01.26, 21:48:17] Daniel: welches modell nutzt du jetzt?
[18.01.26, 21:48:41] Daniel: [clawdbot] Ich nutze gerade Claude Opus 4.5 (über copilot-proxy). Kann ich dir sonst noch was helfen?
[18.01.26, 21:49:12] Daniel: Kannst du mir in 10 Sekunden nochmal schreiben?
[18.01.26, 21:49:21] Daniel: [clawdbot] Ich kann leider keine zeitgesteuerten Nachrichten in Sekunden senden – nur Erinnerungen im Minutenbereich über Cron. Soll ich dir in 1 Minute schreiben?
[18.01.26, 21:49:34] Daniel: Ja schreib mir in 1 Minute
[18.01.26, 21:49:42] Daniel: [clawdbot] Alles klar, ich richte eine Erinnerung für dich ein – du bekommst in 1 Minute eine Nachricht! 🐬
[18.01.26, 21:56:31] Daniel: Ich habe keine Benachrichtigung erhalten
[18.01.26, 21:56:55] Daniel: [clawdbot] Du hast recht, das tut mir leid! Die Erinnerung wurde nicht korrekt erstellt. Soll ich es nochmal versuchen – diesmal eine Erinnerung in 1 Minute?

But I have the following problems:

• he dont save memories
• he dont use skills (whats working for you)
• he only answers when i write first

[sacred swallow]

I installed with npm

[sacred swallow]

I tried to switch the settings in few combinations to see what resolves this problem- even asked clawd to self diagnose and update corresponding settings - but none worked

[warm thorn]

thanks! I mean to give clawdbot the ability to eg login to order my groceries but its running on the vps

Clawdbot set up Browserless and its working - next is to figure out to save my login

But I was wondering if this is the best approach

[sacred swallow]

Create a browser profile , open that browser and login once or save your password and in your skills or prompt ask clawd to use that profile so it doesn’t need to login

I did that with chrome

[floral sand]

Sometimes or I’ll start a Claude code session in tmux get it going the way I want then go out for the day. I tell Clawdbot the tmux name and ask it check in and report back. Works great. We can both control the session as needed. I’m sure there are many other ways to do it.

[floral sand]

That I haven’t done yet sorry. Outside of clawdbot I’ve been able to do things like with Claude in chrome and chrome dev tools

[floral sand]

How do you create a browser profile?

[mild bluff]

Ask your clawdbot to create it and open it and navigate to the right url.

Confirm right profile by clicking the profile(letter?) icon in top right of chrome.

Login.

For this to work you need access to the display of whatever is hosting your bot

[floral sand]

Is there a way to do it if your clawdbot is on a remote vps?

[runic dove]

Has anyone had any difficulties in getting Clawdbot to actually follow your rules and guidelines? I'm trying to get my bot, and we've had many discussions on this, to act as a PM to orchestrate codex coding agents using subagent, to keep the agent:main cleaner.

But it always just reverts to doing the coding itself directly. And it won't set reasoning/models the way I ask it.

It's hard to make it disciplined. Has anyone overcome this>?

[sudden solar]

what model is everyone using to power clawdbot?

[rocky orbit]

GPT5-2 / Codex

[ivory rapids]

Hey @mighty reef - will create a thread here

[wintry bluff]

Anyone getting this windows issue:

PS C:\Program Files\PowerShell\7-preview> iwr -useb https://clawd.bot/install.ps1 | iex

  Clawdbot Installer

[OK] Windows detected
Invoke-Expression: Cannot overwrite variable HOME because it is read-only or constant.
PS C:\Program Files\PowerShell\7-preview>

[mild bluff]

possibly if you tell clawd how to login and you'd have to pass credentials along somehow

[sacred swallow]

Anybody have issue with whatsapp double reply shown in the chat

I solved my earlier notifications problem by giving a new number to clawd

But now the replies are shown twice in my whatsapp chat

The logs and the chat UI on the control panel shows the message only once but for some reason my whatsapp chat with the bot shows the message twice as 2 messages

[small nest]

@flint mica @sly star anything new about github copilot?

[flint mica]

Nah I feel asleep

Switched to my Claude code plan to isolate the issue and make it is a copilot issue

[small nest]

yeah i thought so. claude worked great for me too

[flint mica]

It just uses too much usage

[warm thorn]

Has anybody set up multiple Agents with a Telegram bot each? Been trying for ages and can't get it to work. Only one bot responds 🤔

    "defaults": {
      "model": {
        "primary": "anthropic/claude-opus-4-5"
      },
      "models": {
        "openai-codex/gpt-5.2": {},
        "anthropic/claude-opus-4-5": {
          "alias": "opus"
        }
      },
      "workspace": "/Users/wtan/clawd"
    },
    "list": [
      {
        "id": "main",
        "default": true,
        "name": "Remi",
        "workspace": "/Users/wtan/clawd",
        "agentDir": "/Users/wtan/.clawdbot/agents/main/agent"
      },
      {
        "id": "work",
        "name": "Worker",
        "workspace": "/Users/wtan/clawd-awaken",
        "agentDir": "/Users/wtan/.clawdbot/agents/work/agent"
      }
    ]
  },
  "tools": {},
  "bindings": [
    {
      "agentId": "main",
      "match": {
        "provider": "telegram",
        "accountId": "mainBot"
      }
    },
    {
      "agentId": "work",
      "match": {
        "provider": "telegram",
        "accountId": "awakenBot"
      }
    }
  ],
  "channels": {
    "telegram": {
      "accounts": {
        "mainBot": {
          "name": "RemiClawdBot",
          "botToken": "xxx"
        },
        "awakenBot": {
          "name": "Awaken8_Bot",
          "botToken": "xxx"
        }
      }
    }
  },
  "telegram": {
    "enabled": true,
    "botToken": "xxx"
  },```

[warm thorn]

ah!! 😄 clawdbot figured it out - the docs are outdated:

multi-account Telegram should be configured under telegram.accounts, not channels.telegram.accounts. Here's the correct structure:

"telegram": {
  "enabled": true,
  "dmPolicy": "pairing",
  "groupPolicy": "open",
  "streamMode": "partial",
  "accounts": {
    "mainBot": {
      "name": "RemiClawdBot",
      "botToken": "xxx",
      "dmPolicy": "pairing",
      "groupPolicy": "open",
      "streamMode": "partial"
    },
    "awakenBot": {
      "name": "Awaken8_Bot",
      "botToken": "xxx",
      "dmPolicy": "pairing",
      "groupPolicy": "open",
      "streamMode": "partial"
    }
  }
}

[warm thorn]

cc @low dragon these instructions re channels.telegram are outdated https://docs.clawd.bot/gateway/configuration#channels-telegram-accounts-%2F-channels-discord-accounts-%2F-channels-slack-accounts-%2F-channels-signal-accounts-%2F-channels-imessage-accounts

[rocky orbit]

If you want to pull an update you can!

[small nest]

Yes thats the problem

[floral sand]

Does anybody know how to force your agent to compact memory when autocompact fails?

[urban swan]

Hello, I would like to ask what is the intended workflow of Clawdbot with Claude Code (to build stuff) without enabling YOLO mode? I am trying to make clawdbot manage claude code but it's struggling as it can't see claude code updates and the claude code sessions ends with error code 9.

[mental mica]

are you using the coding-agent skill

[urban swan]

Yes

[mental mica]

have it try exec interactively with pty

https://docs.clawd.bot/tools/exec#exec

[urban swan]

Thank, I will try it

Seems like before it was trying with bash tool, right now it's using exec tool but it's facing same issue - it's probably facing a permission approval prompt which it can't accept as it seems it can't run in background and monitor the claude code with some background agent. It can just interact with the terminal when I message the clawdbot. This I assume is probably not intended behaviour is it?

[urban swan]

It seems like clawdbot is really poor with handling tool like claude code, it's probably not really designed to use tool like claude code right?

[gusty sonnet]

It can remote control claude code quite well. Provided you're on a recentish version and have the coding agent skill. There were recent additions to support exactly this use.

[urban swan]

Well I am on Claude Code v2.1.11; 🦞 Clawdbot 2026.1.16-2 and using with codex model. And it is supposedly using the skill the whole time. It was even having hard time to see permission prompts, it found about it only like 15 minutes later after 7 questions if it sess anything and clawdbot just mentioned claude code is still thinking etc. even though there was permission prompt the whole time.

Could it be one-off occasion then? In this current session?

[gusty sonnet]

Mine seems to handle that permission prompt fine.

[urban swan]

Okay, this seems to work in new session

[terse acorn]

Anyone know why I am not getting any responses from the agent?

[timber ginkgo]

Anyone else running into issues where the web UI isn't properly showing config status? Examples:

1. Yesterday, I was trying to debug some verbose status, and I tried to update in the web ui, but save would never become available for me to click. I had to edit things manually, and only then did it start actually showing up properly.
2. My heartbeat commands in HEARTBEAT.MD aren't being respected, and when I asked it "why" it said "It’s in the system prompt under the Heartbeats section" which makes me believe it's just straight up not listening to what I asked it to do?
3. I'm trying to change thinking levels, but once again, save isn't clickable -- I was able to change it in the TUI, however, once again, it doesn't reflect in the web UI. 😭

Unclear how I even got into this scenario? I installed via NPM.

[keen hollow]

Has anyone found success with Claude Code CLI as their model

Tried for 2 days, nothing

Connected to Claude API, instant success

But burns so quickly

[mental mica]

thats how @silk ingot and @sharp parcel and my clawdbot all run

[keen hollow]

As in burn quickly?

Wondering if I should just use somehting else as model

[vivid thistle]

I'm getting an issue where any clawdbot command just hangs, I can't even exit it. Anyone else encounter this? I am on the latest commit

Edit: Running any clawdbot command hangs indefinitely and uses 100% of the CPU core

[random spoke]

@vivid thistle @woven blaze

I've been investigating similar. I think there's a bug compounding between the new hard failure when there's a non-existent property in the clawdbot config and the recent removal of the bridge protocol. All my clawdbot xyz were getting into a Invalid config: - : Unrecognized key: "bridge"reboot loop and ramping to 100% cpu. Try the clawdbot doctor --fix but if that's hanging too (mine was) maybe check your config file and remove the bridge config if present. Mine came back up once I manually removed it.

[mental mica]

worth a try on this pr

https://github.com/clawdbot/clawdbot/pull/1272

[woven blaze]

doctor fixed work. maybe we should ahve doctor run automacically?

[mental mica]

https://docs.clawd.bot/install/updating#always-run:-clawdbot-doctor

[neat iron]

do yall still have any commands related to plugins be very slow tho? like if you run clawdbot doctor it hangs on on the skill status section which means it's running something related to plugins (next section)

[thick hazel]

Anyone get a permission error when trying to use the slash command for 1Password?

[lethal prawn]

hi guys, is there anyway for clawd to control a warp tab

[vivid thistle]

Except I can't even run doctor, even running clawdbot --help hangs

[misty light]

why do i have to grant permission every time now that clawdbot runs a CLI command on my mac

[modest estuary]

I have the same issue

Installed through source and using pnpm

Everything hangs, multiple clawdbot processes starting up, 100% cpu usage, nothing in the logs

Cloned the repo again, threw everything away except for ~/.clawdbot and the workspace

Same issue

No problems at all if not installed from source....

I also see this when running something like clawdbot tui, again 100% cpu for hours [clawdbot] Building TypeScript (dist is stale).

[floral sand]

Anybody else having serious issues with compaction and autocompaction?

Autocompact fails to trigger automatically (I think it worked once). Manual compact works sometimes fails sometimes. Token counts shown in /status are not accurate. Compact count is also totally wrong.

(⚙️ Compaction failed: Summarization failed: 400 {"type":"error","error":{"type":"invalid_request_error","message":"invalid params, context window exceeds limit (2013)"},"request_id":"05bdc13...e11cb6"} • Context 185k/200k (92%))

[limber rapids]

hey @midnight fox you run into this? on slack. duplicate messages. I don't wanna turn socket off since I want them to recieve messages in real time

[limber rapids]

*socket off

[midnight fox]

I have not used slack for a week tbf, im having a lot of joy with whatsapp and spawning of agents.

However, I will likely be trying it out in a couple of days again.

I did see someone on X saying they had done slack to use webhooks and it was much quicker but havent tried

[limber rapids]

yes i wanna find that now, i saw it too. please point me to it if you see it first.

I'm a telegram guy with my agents, but i need to bring my team into the light, and they are on slack.

[limber rapids]

@rustic sigil I believe this was you a few days ago?

Got it https://github.com/clawdbot/clawdbot/pull/1143

[drifting helm]

What's the latest guidance on models / plans to use? Opus via API definitely expensive... Minimax 2.1 is also similarly expensive in my experience; while its list price is lower, in practice I have a lot more follow up with it. Are folks using a minimax plan somehow?

[undone ice]

Hi everybody, quick question: I just setup many thing with the cli, but I now saw that there is a Mac app that might have some interesting permissions (record screen and audio). If I now install the Mac app, is it two different instances? Can I transfer my setup or do I have to redo everything?

[mental mica]

It can remote connect over Tailscale to your main gateway

[undone ice]

It would be on the same machine so tailscale wouldn’t be needed right?

[small nest]

@flint mica What do you think about an MCP for GH Copilot?

working on something..

[vivid thistle]

I fixed it last night forgot to mention it here. Looking at the logs, it turned out the config file had keys that it didn't recognize. It looks like the newer versions changed how the config file is structured.

I renamed the clawdbot.json file to back it up, then ran clawdbot configure to regenerate the file.

The fact that an invalid config file causes an infinite loop seems like a bug on their end though

[modest estuary]

I noticed the key error as well. It's great to see you explored that issue further. Yes it was introduced a couple of days ago, didn't happen before. Thanks for reporting back!

[green oak]

I've installed this yesterday and today had it working on a browser automation task, and then it suddenly just stopped. Now I can't get it to respond via telegram or the local UI, I see this in the log:
{"0":"{"subsystem":"agent/embedded"}","1":"Auto-compaction failed (unhandled): Context overflow: Summarization failed: 400 {"type":"error","error":{"type":"invalid_request_error","message":"prompt is too long: 206425 tokens > 200000 maximum"},"request_id":"req_011CXK1PrVbaZNCmocUr9BoC"}. Your input may be too large for the context window.","_meta":{"runtime":"node","runtimeVersion":"22.22.0","hostname":"unknown","name":"{"subsystem":"agent/embedded"}","parentNames":["clawdbot"],"date":"2026-01-20T17:15:36.257Z","logLevelId":5,"logLevelName":"ERROR","path":{"fullFilePath":"file:///home/[path]/.nvm/versions/node/v22.22.0/lib/node_modules/clawdbot/dist/logging/subsystem.js:150:16","fileName":"subsystem.js","fileNameWithLine":"subsystem.js:150","fileColumn":"16","fileLine":"150","filePath":".nvm/versions/node/v22.22.0/lib/node_modules/clawdbot/dist/logging/subsystem.js","filePathWithLine":".nvm/versions/node/v22.22.0/lib/node_modules/clawdbot/dist/logging/subsystem.js:150","method":"logToFile"}},"time":"2026-01-20T17:15:36.257Z"}

[green oak]

I think I just ran into this, how did you fix it?

[green oak]

ok if anyone else runs into this, it's apparently a known issue: https://github.com/clawdbot/clawdbot/issues/1280

I fixed it by deleting a bunch of stuff out of the session file, which I found with the command on that issue page: find ~/.clawdbot/agents/main/sessions -name "*.jsonl" -size +200k

[floral sand]

I don' t think I did. Still trying to figure it out.

Of course you can always just start a new session. But in terms of rescuing the problem session what you did or something similar is all I know of.

[torn compass]

i sent around 10-15 messages and it already used $10 of opus usage. is it just the context window becoming full? are you all using a different model? or is this just how expensive it is

for context, I had asked clawdbot to read a lot of files in my obsidian vault.

[modest estuary]

It seems normal to me. Perhaps the heartbeats are adding up too

[graceful mauve]

the hetzner instructions are a mess

[molten pond]

just FYI this is still happening

[inland forum]

I setup Signal to allow from two phone numbers and then created a group chat on each with my clawdbot number. Clawdbot will only chat with one of them, the other one does not come up in the dashboard how do I get this working?

[sacred swallow]

I think it’s fixed for me

Was happening couple of days and now stopped

[topaz pike]

hi, aside from giving ralph (ralph loop) a PRD, which I did as well, how do any of you manage the session limits, I seem to keep hitting them, and claude took a good while to get it to run. I am on the pro max plan.

[craggy sorrel]

I've tried using the #1459642797895319552 channel but the results weren't particularly useful. I'm running into an issue with the iMessage gateway. I just installed it from scratch on a new machine and it seems to not be able to use imsg as expected. (To be clear I had a real phone number, I just redacted it for the screenshot.)

[shell patrol]

did the provider setup options with oauth (not api key) go away? (or did I imagine those in the first place)

[frozen bolt]

How can I turn a rasberry pi into a node? I can run it and connect to the remote gateway (external VPS) but can't get it visible as a node 🙁

[tidal fox]

Are people using their Claude subs without issue or getting banned ?

[tidal fox]

Run the node host on your Pi and point it at your VPS bridge port 18790 (not 18789):

clawdbot node start --host <VPS> --port 18790

Then on the VPS approve it:

clawdbot nodes pending → clawdbot nodes approve <id> → clawdbot nodes list

[tidal fox]

Clawdbot onboard again and should see it?

[tidal fox]

Heartbeats def affect context I reduced mine.

[torn compass]

oh what is a heartbeat?

[tidal fox]

https://docs.clawd.bot/gateway/heartbeat#heartbeat

I had a lot of success just pointing my clawdbot to www.clawd.bot and said search about this etc. or leverage ChatGPT or Claude or Claude code for set up optimizing it and just keep saying. Double check the https://docs.clawd.bot/getting-started link and https://github.com/clawdbot/clawdbot

[vital island]

Anybody successfully able to wire up GLM 5-7-Flash? I’m using Zai but not working on ClawdBot for me.

[fickle turret]

This is the way to use your Claude Max subscription on Clawd while also still using Claude Code CLI on the same machine

Intial setup

When run claude setup-token, after you do the auth on the browser, it returns to claude code. You need SCROLL UP and read the stuff.

There is an export command you need to run in your terminal to make the long-lived token accessible to Clawd. Then, you need to go back to the clawd onboard wizard and paste the CLI token into clawd (option 3).

Problem: Claude Code CLI + Clawd share OAuth token → using Clawd logs you out of Claude Code CLI

** Fix:**

bash

1. Create separate config dir
   mkdir -p ~/.claude-clawdbot

2. Login again in Claude Code CLI (opens browser)
   CLAUDE_CONFIG_DIR=~/.claude-clawdbot claude

Type /login, authorize, exit

3. Make permanent
   echo 'export CLAUDE_CONFIG_DIR="$HOME/.claude-clawdbot"' >> ~/.zshrc
   source ~/.zshrc
   Clawdbot users: Add "CLAUDE_CONFIG_DIR": "/path/to/.claude-clawdbot" to env.vars in
   ~/.clawdbot/clawdbot.json.

[knotty onyx]

do you use git checkout version or global install on your remote server?

[upbeat basin]

Have seen people getting banned for doing this though - proceed with caution

[opaque swan]

hey, are there any obvious differences between using cc and clwd as a “personal assistant”?

if i’ve understood correctly cc is sub based is api and clwd has quite a lot of purpose built functionality for the role

[ebon flare]

I look forward to start giving Clawdbot access to things like my social media accounts and email. But I notice I'm doubting a lot because of prompt injection risks. Curious to hear if others feel the same and maybe if there are any tips on how to address this (besides of using Opus 4.5 as a model)

[marble flume]

**why the heck would it no longer know the command /dock_telegram, how could clawdbot lose this type of information? **

I don't have information about the /dock_telegram command in my current context. It sounds like it might be a Clawdbot-specific command, but I don't see it documented in the CLI reference I have access to.

To find out what it does, you could try:

clawdbot help to see all available commands
clawdbot dock_telegram --help if it's a subcommand
Or let me know if you'd like me to search through any documentation files that might contain this information

You're right to question that! I should clarify - I have access to information about Clawdbot's tools and some CLI commands, but my knowledge isn't necessarily complete about every command that exists.

Looking at my available context, I can see:

Various tools (exec, browser, message, etc.)
Some CLI commands like clawdbot daemon start/stop/restart
Skills for different integrations
But /dock_telegram isn't mentioned in any of the documentation I currently have access to. It's possible:

It's a command that exists but isn't documented in my current context files
It's from a different version or configuration
It might not be a real command (you could be testing my knowledge)
Rather than guess or make something up, I should be honest about the limits of what I can see in my current context. If this is indeed a core command I should know about, could you share what it does or where I should look for documentation about it?

[signal tartan]

Hi,
has anyone getting clawdbot as daemon on a mac running with iMessage over imsg? Clawdbot need FullDiscAccess (FDA) for accessing chat.db. In the terminal with "clawdbot gateway" ist no problem to give the terminal FDA. Clawdbot as daemon start node itself. I tried a wrapper app to give node FDA, because you cannot add a binary manual FDA. But this does not work. Any solutions?
V2026.1.20, Mac Mini M4, macOS 26.2

[signal tartan]

Clawd need FDA to access the chat.db from iMessage. When you install clawd, clawd starts the gateway as a daemon. This use the node directly, but you cannot give a binary like node manual FDA over Settings > Privay > FullDiscAccess.
I've the same issue, but no solution.

[mental mica]

your bot doesn't automatically know everything about itself, clone the codebase and let it look through it

[signal tartan]

I tried a wrapper app that pointed to the node path to bypass the the node and give the wrapper app FDA over Settings > Privacy. But this didn't work.
In the first time i installed clawd in a vm and a wrapper app worked. But now on my mac mini, it does not. I don't no how i got it work in the VM 😫

[storm saddle]

Anyone else ever run into │ Telegram │ WARN │ failed (unknown) - This operation was aborted ? this started happening out of the blue today 🤔

[signal tartan]

Has anyone this Log-Message in 2026.1.20-2: 17:00:12 [tools] exec failed: exec denied: approval required (approval UI not available)? In 2026.1.16-2 no issues.
macOS 26.2, Gateway in iTerminal with "clawdbot gateway --verbose" (not daemon), OpenAI Codex 5.2

[wintry ibex]

It seems very fragile. I got it working and then it broke.

[willow scroll]

noob trying to setup clawdbot and stuck in an infinite loop, any chance someone can get on a sharescreen with me and help me out of my misery ?

[mental mica]

Have you tried #1459642797895319552 yet

[willow scroll]

extensively

[mental mica]

I just found it, I don’t have any experience with Jim and I and I’m pretty sure that’s where your issue is

LOL

Gemini

[willow scroll]

so you think switching to anthropic will make things easier ?

[floral sand]

Gemini seems to be broken in Clawdbot. Claude works great and so does MiniMax M2.1

[scenic osprey]

Hey there - anyone had luck having an agent set "timers" or "reminders"? I feel like I'm really close, but my agent keeps getting The gateway still shows "gateway closed" for cron operations even though I've clawdbot config set cron.enabled true and restarted the daemon several times. I imagine the agents are able to set crons, right?

[obsidian orchid]

Question for y'all: when I set up my Clawdbot like a week ago, Anthropic had very recently disallowed the ability to use Clawdbot with Opus via just a Claude Max subscription, it wasn't working at all. So I set it up with API calls to my Google/Gemini account.

Now I see people talking on twitter as if that has changed again, and it works again just to use Clawdbot with Claude Max/Opus. Is that right or am I going crazy?

Is this just OAuth loophole vs setup-token approach maybe?

[shell patrol]

I tried the new curl&bash script installer and it had only API key options.

[lofty spade]

Hi guys, what is the best way to install Claude on a generic Linux VPS? I see these repos about ansible. I'm not familiar with that, but is that the best way? Given that it's out there on the internet, i'm concerned about making sure it's locked down properly. All advice appreciated.

[mental mica]

install script on https://clawd.bot

then just lock down your server using any guide on the internet or have your clawdbot help you

[wide cove]

Hey all! Loving Clawdbot but just updated from .16 to .20-2 and am now receiving an error: unknown command '/usr/bin/node-22' when trying to run the gateway? I don't see any mentions of it when searching Discord so it might be something unique I've fudged up?

🦞 Clawdbot 2026.1.20-2 (cb7791c) — I speak fluent bash, mild sarcasm, and aggressive tab-completion energy.
error: unknown command '/usr/bin/node-22'

│ Version │ 2026.1.20-2
│ OS │ linux 6.12.0-124.20.1.el10_1.aarch64 (arm64)
│ Node │ 22.19.0
│ Config │ /root/.clawdbot/clawdbot.json
│ Dashboard │ http://127.0.0.1:18789/
│ Tailscale │ off · Running · xxxxxxxxxx.tailbda818.ts.net
│ Channel │ stable (default)
│ Update │ pnpm · npm latest 2026.1.20-2
│ Gateway │ local · ws://127.0.0.1:18789 (local loopback) · unreachable (connect failed: connect ECONNREFUSED 127.0.0.1:18789)
│ Security │ Run: clawdbot security audit --deep
│ Gateway self │ unknown
│ Gateway service │ systemd installed · enabled · running (pid 2066818)
│ Node service │ systemd not installed
│ Agents │ 2 total · 1 bootstrapping · 0 active · 5 sessions

[dire spade]

I couldn’t even get it to accept my subscription. Would consistently get auth errors and only works when with console api key

[tidal fox]

Hmm explain why it’s saying my token is old and I cannot update it via oauth

[wide cove]

Rolling back to 2026.1.16-2 seemed to get rid of the issue. Sounds like a bug that @silk ingot might need to be aware of??

[inland forum]

If I want to setup ClawdBot for a small team on Linux, what is the correct setup to allow all team members to access the main/shared workspace but allow each member to have a private workspace that has read access to the main/shared memory?

[grim ruin]

Hey everyone, non-technical noob here. I've bought a mac mini to use as a server, installed clawdbot on it, and now I'm trying to setup my macbook as a node to access the mac mini server. I'm having trouble trying to connect. I've tried using local ssh, using tailscale etc. Whenever I try to change the SSH address it keeps defaulting back to jordan@, which is the username for my macbook. I think I need to change this to point to my mac mini but it's not allowing me to. Any help would be greatly appreciated 🙂

[limber cape]

whats the best model to run as backup when claude max limit gets hit

[rocky orbit]

Did you try #1459642797895319552 ? 🙂

[signal tartan]

I running gemini over the gemini-cli with OAuth to my account. Works great!

[signal tartan]

Do you ask the agent to use the „cron“ tool?

[signal tartan]

The v20 has a new access managment. Do you read https://docs.clawd.bot/tools/exec-approvals ?

[molten pond]

It's definitely still happening ... aggressively.
Clawd has reached a place where it's commenting on its own heartbeats 😂

[wide cove]

I did set this up but the gateway would still crash immediately - clawdbot status, clawdbot onboard etc all worked fine but the daemon/gateway commands always gave a node 22 error despite being all set up. I’ll try again today now I’ve had some sleep!

[gray light]

Hey, with gemini cli with oauth and I get error: loadcodeassist failed: 403 forbidden.
I have put in export CLAWBOT_GEMINI_OAUTH_CLIENT_ID as well session secret before running the auth. Anything I’m missing?

[signal tartan]

Jump into gemini-cli and Make Autorisation Inside

[grim ruin]

Yeah I did, unfortunately it hasn't been resolved yet.

[leaden tinsel]

https://youtu.be/5kkIJNUGFho?si=F5D0E_HL8CSkUGCc

[gray light]

Sorry I dont get it what do you mean? I authorize from gemini cli inside the vps where clawdbot is?

[willow scroll]

thanks, i'll try !

[meager mortar]

when do people use Claude Code over Clawdbot (for non-coding tasks)? I feel like Clawdbot chews up a lot of tokens (conspiracy theory: Anthropic charges Claude Code fewer tokens then 3rd party access). So lets say i want to organize some markdown files -- clawdbot (Opus) or claude code?

[lofty spade]

should i not be using ansible? is this not the "best" way to install clawd?

what are the pros and cons?

[floral sand]

There is a GitHub issue on it if you want to follow it: https://github.com/clawdbot/clawdbot/issues/1305

[unkempt bridge]

No matter how I install clawdbot on a clean Ubuntu live server, it fails to actually use and install skills from clawdhub. No matter which skill I ask it to install, it tells me either that it can’t see the skills, or that it can’t see the requirements that the skill needs.. what am I doing wrong??

[bitter jetty]

running into an issue this morning with /compact etc not working, context just keeps growing until I manually delete the session files. anyone else?

[bitter jetty]

I got /reset to work by sending it from Telegram instead of the Web UI interestingly enough, still no luck on /compact for some reason

[floral sand]

It's been buggy and inconsistent for me too... but I also noticed that the compact counter doesn't always work. Sometimes it displays 0 but successful compacts did take place.

[vestal coral]

Hello Hello. I have been struggling to set up the Gmail integration and after about a day or so of fighting with various Google OAuth and GCP permission issues i am now connected. However the watch server keeps running into issues that the port is already taken.

Has anyone here had the same issue and maybe found a way to resolve it?

watch: listening on 127.0.0.1:8788/ listen tcp 127.0.0.1:8788: bind: address already in use gog watch serve exited; restarting in 2s

This is what I get over and over again. Killed the previous process but doesn't really help.

[civic harness]

Hello, I try to connect Clawdbot with Claude Code CLI but I just can't, I run claude configure and have this type of error (image)

you can see this thread for more details https://discord.com/channels/1456350064065904867/1463937907856244758

[mental mica]

This message says it’s trying to use keychain mode not the setup-token mode

[civic harness]

Is this not this option to connect Claude Code?

[mental mica]

I think that’s outdated, you need to update, but also you want to use the setup token option below, not the cli

[civic harness]

Oh okay weird, I've installed Clawdbot yesterday

This is driving me crazy 😂

[signal tartan]

I’m running Gemini CLI locally on my Mac, not directly on a VPS.

Gemini CLI authenticates via Google Account login. When you start it, it opens a browser window to sign in, and after that it uses the account-based quota automatically. Cloudbot itself is only configured with the model name and just talks to Gemini CLI.

On a headless VPS this is likely the problem: the login flow expects a browser, which usually isn’t available. That’s why it works locally for me, but may not work on a terminal-only server unless you add some browser/OAuth workaround.

[civic harness]

And now I just can't start clawdbot since update, I give up

[sage imp]

you just want to use your existing claude code? i think you can just callit: https://docs.clawd.bot/gateway/cli-backends#beginner-friendly-quick-start

[signal tartan]

"contextPruning": {
"mode": "off"
}

[sage imp]

this is setting up which model you want clawd to use, not connecting claude code as a tool

[civic harness]

Hmm okay, so how can I connect my Claude Code to clawdbot?

[signal tartan]

If you want it like before 2026.1.21

[sage imp]

did you check out the previous link? you just want to use your existing claude code? i think you can just callit: https://docs.clawd.bot/gateway/cli-backends#beginner-friendly-quick-start

[civic harness]

Doesn't works

[civic harness]

Yes I've checked, I don't see how this relates to my problem, sorry. What command are you referring to exactly?

[sage imp]

you're trying to use your existing claude code installation, yes? or something different

[civic harness]

I'm trying to use my claude code subscription through clawdbot

[sage imp]

to run the bot or use clawd to use claude code

[civic harness]

to run the bot

to discuss with clawdbot and use my claude code subscription to not use api usage

[sage imp]

try the token method instead if you're not having luck. that worked for me for my anthropic pro subscription

[civic harness]

hmm hmm, I've tried both, both doesn't works

[signal tartan]

mode: off works for me, after i updated to 21, but i dont use claude. Sorry.

[sage imp]

i see you have a max sub from your help thread. skip claud code altogether and choose opus and the token mode

[granite onyx]

For clawdbot Codex vs Claude which one you are using and why ?

[civic harness]

Yes I would love to do that but this doesn’t works too 😂

[sage imp]

where does it breakdown?

[rocky orbit]

Codex because it is more generous with rates

[stark gyro]

hiya, has anyone seen cases where the clawdbot browser relay detaches after some X amount of time? i've noticed that i can attach it at night but when I wake up even though I don't think my server has gone to sleep, it will have detached, which makes some of my cronjobs fail due to lack of browser access

[vestal coral]

Hey honest question, has anyone succesfully connected clawdbot to gmail?

[rocky orbit]

gogcli 🙂

[vestal coral]

yeah I have that part, just that clawdbot webhooks gmail run wants to run on the same port as gog, which means it can never run.

[flint mica]

@small nest any updates on the copilot proxy issues?

[sage imp]

anyone else having issues post update with antropic auth on linux? (VPS)

[flint mica]

What’s the issue you’re having?

[mental mica]

#old-models we’ve seen lots of Anthropic issues today

[sage imp]

Token authentication, re did wizard, pulled new token, removed and re-created several JSON files per krill. Log says no API key for anthropic, model status throws "TypeError: Cannot read properties of undefined (reading 'trim')"

krill threw up its hands and said to create a bug report

seemed to be working fine before i pulled the latest so timing but haven't been able to get any signs of life

[vestal coral]

how do y'all connect to anthropic? the bot ate through my $10 api budget in about 5 minutes of mild use.

[wicked jewel]

Multi-account Telegram: webhooks + heartbeat routing?

I have two Telegram bots - a main bot and a secondary "heartbeat" bot. I want heartbeats handled by a dedicated heartbeat agent and sent via the heartbeat bot.

I set up multi-account using channels.telegram.accounts with a binding to route the secondary account to my heartbeat agent:

{
channels: {
telegram: {
accounts: {
default: { /* main bot */ },
heartbeat: { botToken: "...", webhookUrl: "" }
}
}
},
bindings: [
{ agentId: "heartbeat", match: { channel: "telegram", accountId: "heartbeat" } }
]
}

Two issues:

1. Webhooks: When I enable webhooks for both accounts, I get EADDRINUSE on port 8787. I set webhookUrl: "" on the secondary account to force polling as a workaround, but this adds ~60s latency.

2. Heartbeat delivery: With target: "telegram" in the heartbeat config, messages send from the default account. I couldn't find an account field in the heartbeat schema to specify which bot sends.

Is the channels.telegram.accounts + bindings approach correct for this use case? I saw a Discord thread mentioning per-agent telegram config (agents.paul.telegram.token) - would that pattern work better here and solve both issues?
per-account telegram config thread
#old-help message

[gray light]

anyone having issues with tool calls recently? I am on the git install method and the agent is outputting tool calls on telegram like : <ls>
<path>/root/clawdbot/extensions/lobster</path>
</ls>

[small nest]

Sadly no. Was still the same fails like before. No memories, no tool handling.

Tried the same with perplexity, but this has the same issues (still with API)

The only idea I had was using a custom MCP agent. Tried building one with ki, but always my tokens run out to try something more.

[surreal zenith]

I got the same error after update. I deleted contextpruning.mode outright, restarted gateway daemon and never looked back.

[amber crypt]

same issue

[amber crypt]

this fixed it. But for a few hours (pre and post update) I havent been able to start a gateway...

[undone ice]

same! Also when using skills. Not sure what caused it yet, I am investigating. Did you manage to solve it?

[gray light]

Yeah deleting the lobster plugin as well as delete agents tools in clawdbot.json solved it for me!

[undone ice]

This solved it as well for me! Thank you 🙏

[split grove]

hey folks, good morning. I have some huge issues to let clawdbot respond or make a task in the specific date and time on macos. How do you do it?

it keeps forgetting DST, tries to use crontab repeatably (which is no-go on macos) etc

[floral sand]

Did anyone definitively figure out the answer to huge token burn with Claude (subscriptions, not API)? I have switched to MiniMax for Clawdbot while I figure it out but I can see that quite heavy use outside of Clawdbot only reaches ~50% of the $200 plan limit, while when I was using it with Clawdbot I burned up the whole week's quota in 2 1/2 days.

[bright oxide]

I think I'm still using the old Anthropic OAuth. Should I change that to the CLI token mode or am I fine running the old mode as it is?

[gusty sonnet]

There were some recent updates to how clawdbot handles caching that should improve token burn.

[floral sand]

I thought those only effect Antrhopic API?

[tough crater]

has anyone built their own plugin and had issues with outbound configs?

[mighty needle]

anyone get this when using vps - says i need to reresh auth every 8 hours? ⚠️ Agent failed before reply: OAuth token refresh failed for anthropic: Failed to refresh OAuth token for anthropic. Please try again or re-authenticate.. Check gateway logs for details.

[novel hedge]

Prompt caching works similarly regardless, so see if it helps you

[pale shard]

Hey @gray light , have you managed to get passed the 403 forbidden on Gemini cli oauth ?
Thanks 🙏

[dense glen]

Ok, I'm getting a bit disillusioned. And I don't want to. As an example - I want it to check a folder of my emails and summarize things every X hours. Then I want it to be able to mark them as read if I say so. My no-longer-trusty clawdbot set up a heartbeat. After 5 attempts, it managed to actually summarize on a heartbeat. It asks "mark as read?" and I say yes. And it has no clue what the 'yes' is for. Is this.. normal? haha

Am I just misusing it?

[mental mica]

The heartbeat happens in the main session, are you in the main session?
Also what model

[dense glen]

How do I know what is the main session? And Opus

I kind of don't think it is the main session, though..?

[keen hollow]

with my claude max plan. The burn is much lower than api % wise

[keen hollow]

I had the same problem and must've spent at least 12 hours to get it running initially. Had my clawd build a skill that reads keychain + edits the auth profile config and automatically refreshes before expiry for me.

https://github.com/TunaIssaCoding/claude-oauth-refresher

I tested it myself and it works, think it might be helpful for your situation

[mental mica]

Correct, main agent but not main session

[dense glen]

My clawd said it can set the session to that for heartbeats. I'm gonna try that. Thanks for the hints!

[civic harness]

Wow I will try this thank you very much

[keen hollow]

let me know if it doesnt work! I'll try to work on fixes

[keen hollow]

I'll be back

[keen hollow]

npm update -g clawdhub

Can you try running this, then run the install again

[civic harness]

same issue

[keen hollow]

cd ~/clawd/skills
git clone https://github.com/TunaIssaCoding/claude-oauth-refresher.git
cd claude-oauth-refresher
./install.sh

Try this

[civic harness]

@keen hollow I'm connected to Claude Code

[keen hollow]

Have you used

clawdbot config

to set up model for the first time?

[civic harness]

yup

this doesn't works

[keen hollow]

damn ok. Let me try and replicate on my other device

I'll be back

[civic harness]

I have this error since yesterday, not linked with my initial problem

[keen hollow]

Ah. You need the pruning mode off thing. I had this too. I think it's problem with the last clawdbot version

I know you tried turning it to off earlier and it didn't work right

[civic harness]

Yes with the config, how do you done that?

[keen hollow]

lemme try and see

[civic harness]

How do you have fixed this?

[keen hollow]

1. run "clawdbot models"
2. read from top, the first row says Config, followed by a path of your clawdbot.json file.
3. Copy that path
4. run "nano [paste the path]"
5. scroll down until you find " workspace"
6. the row under should say

"contextPruning": {
"mode": "[whatever you have right now]"

7. turn [whatever you have right now] to "off"
8. control + O to save, control + X to exit

Then try the refresher again

@civic harness

[civic harness]

nice @keen hollow everthing seems to works, I just can't open the web GUI I don"t know why. My gateway is started

[somber wedge]

IS anyone able to help me with a challenge I'm experiencing?

[undone tinsel]

What model have people generally been using that isn't ridiculously expensive monthly?

[civic harness]

hmm nope, this is not working

[keen hollow]

Ok. I’m going to bed but I’ll try to replicate and solve this first thing in the morning

ETA 10 hours

[civic harness]

yes no problems! thank you

have a good night 🙂

[keen hollow]

Thanks brother

[somber wedge]

Is anyone able to help me think through this?

[keen hollow]

@civic harness the CLI is installed, but have you tried

• run “Claude”
• after the Claude code cli pops up, run “login” and authorize from your account (a webpage would pop up asking for permission)

[civic harness]

yes

[keen hollow]

Can you check your keychain access, type in Claude, double click Claude code credentials and click show password

The password tab should show the token

I imagine you ran that before right

I think earlier you said you used paste token

If you can’t find Claude code credentials in your keychain, you must log in on CLI again

I think they auto delete after a period if not used or something

[civic harness]

I have it yes

[keen hollow]

And in the show password section, does it contain access token and refresh token

[brazen crater]

how are most folks doing updates? via clawdbot itself or command line?

[civic harness]

It contains a string, I don't know if it is the refresh or the access token

[toxic rose]

Recently claude code banned opencode from using their tokens. Does clawdbot also come under this?

[keen hollow]

If you copy the whole thing you should see it say “access token:….” “Refresh token:….”

If not it is the wrong file and you need to log in again

[civic harness]

I think this is the wrong file. It is weird since I've logged in to Claude Code a dozen of times since yersterday

[keen hollow]

YES

awesome

I have the same issue before

[civic harness]

with the /login

[keen hollow]

I’m not sure who but a second Claude code credentials without the needed information is created

Just login again and type “Claude” in keychain again

And you should see two keys present at the same time

You don’t need to do anything in keychain once the second pair of key appears

[civic harness]

Yes exact

[keen hollow]

The refresher will read the correct one

Now try running the refresher again

[civic harness]

nope

[keen hollow]

😢

[civic harness]

😂

[keen hollow]

I deleted my Claude code cli

Lemme try and run it

[wicked jewel]

Does anyone here have Telegram working webhooks (vs polling) with two bots mapping to two agents?

If so, can they explain the setup? I tried getting this working and debugging it with Claude Code for hours and ended up with one having to be on polling and one on webhooks. And even then, I couldn't find the way to have heartbeat responses sent to the heartbeat bot.

Maybe this setup is just not supported. Does anyone have any insight?

[keen hollow]

Ok I've recreated the problems

Now to solve it

[severe rampart]

I am unable to select OpenRouter models. It throws the error - Error: Unknown model: openrouter/auto. And the config file looks like this - Any help will be appreciated.

[keen hollow]

ok @civic harness

try this first

[severe rampart]

nvm, after searching the discord I see that openrouter/google/gemini-3-flash-preview works! maybe this is a bug in clawdbot configure that should be fixed.

[keen hollow]

1. clawdbot config
2. in models config, select anthropic > anthropic token (paste setup-token)
3. manually copy and paste the string after "access token" but before "refresh token" into the config prompt

Then try clawdbot models

not claude code cli

that consumers

consumes

Do exactly this, nothing in between, and in "clawdbot models", at the bottom you should see

ah

I think I found the issue

let me try and fix it

[wicked jewel]

I don't know how openrouter/auto works, but the docs suggest a naming pattern like this: openrouter/anthropic/claude-sonnet-4-5
See https://docs.clawd.bot/providers/openrouter

[keen hollow]

@civic harness what does the "account" section of your keychain say

the correct keychain

is it your username

@civic harness it's fixed

[mystic ginkgo]

does anyone have recommendations for free fallbacks? i was trying openrouter but seem to be getting rate limited

[keen hollow]

Have deleted the old one and uploaded under a new name for cleaner process

https://clawdhub.com/TunaIssaCoding/claude-connect

Just follow the steps and it'll work

[granite galleon]

i keep getting this error when trying to setup the matrix plugin:

- plugins.entries.matrix: plugin matrix: duplicate plugin id detected; later plugin may  │
│    be overridden                                                                          │
│    (/home/clawd/.npm-global/lib/node_modules/clawdbot/extensions/matrix/index.ts)

does anybody have anyidea why is this happening?

[modest sphinx]

is anyone using local models? wondering which models have worked best for folks. i keep running into rate limits w/ claude (set-token) and gemini (api key), so i'm considering spinning up a local model.

[mystic ginkgo]

im wondering the same thing. i set up ollama but im getting a warning about using small models, which require sandboxing.

is per model/provider sandboxing possible?

clawdbot did this but not sure if this is a real or hallucinated config:

  "agents": {
      "models": {
        "ollama/qwen2.5:7b": {
          "alias": "qwen",
          "params": {
            "sandbox": {
              "mode": "all",
              "workspaceAccess": "ro"

[granite galleon]

SOTA is currently GLM-4.7 flash

[daring siren]

Is there a guide on how I can set up clawdbot on my vps and use it with codex with my chatgpt subscription ?

[untold kestrel]

does anyones clawdbot say its doing something and then it just never reports back with the result? current example was i asked it to show me the contents of a markdonwn file and it says "im accessing the file to display to the user" it then never showed me so several minutes later I asked what it said and then it gave me the answer immediately that time. it seems to do this all the time for me

[fallen bluff]

I realized something is wrong with my context pruning / purging config once i saw every message was taking ~5% of my Codex usage. I can't delete my main session, how can i manually prune so i can start talking to clawd again? (on v 1.22)

[gusty sonnet]

Yes, it gets stuck waiting for somethign and you have to poke it to get it moving again.

[compact pawn]

Anyone have clawd as an orchestrator for claude code agents? Any guides on setting it up (is there smth special i have to do other than just installing/authenticating w claude code)?

[mental mica]

there's a coding-agent skill built in

[compact pawn]

I see, thanks. Seems like claude headless mode is consistently erroring for me with ````tool_use` ids must be unique```

any chance anyone else has run into this?

[unreal hedge]

when I send an image. Is there a reason it cannot see the attachment of an image?

[mental mica]

thats your model

[unreal hedge]

Opus 4-5 cannot see images?

[cunning quest]

Help any1? Tried, reset, restart, new gateway etc
LLM request rejected: invalid params, tool result's tool id(call_function_27mogzok5e2l_1) not found (2013)

[wispy tree]

hey guys, ive been struggling with an issue all day when trying to connect claude to my clawd bot, it seems to be struggling with both the setup token and the CLI - according to mty claude, it says that anthropic changed their usage to only be API for things that are not claude code. is this right or suer error?

[plain light]

@wispy tree you probably need to update clawdbot not sure if you have an old version. This was an issue a few weeks ago and there were some fixes posted

[wispy tree]

strange, I just installed it but ill try to force an update

[sinful dawn]

My clawdbot says its going to start a job, and just doesn't. originally i thought it was dying on tool calls (waiting on permissions for one) but now that thats fixed, its still doing it.

weird right?

[keen hollow]

@wispy tree can use what I made here to fix that

It detects your keychain and fills it into config directly

[wispy tree]

ill try rn

[keen hollow]

Do make sure you login on your Claude cli before using the refresh function just like the docs says.

Clawdbot wizard eats the keychain during set up

[wispy tree]

Still not working for me, still getting 401 errors in TG with authentication errors

Even after running all of that

been at this for about 7 hours now

might break my mac mini lol

[sinful dawn]

just keeps doing it. I'm going to dig into it.

[keen hollow]

Can you show me what returns after the you ran the refresh function

That way I’d know if it’s a connecting model problem or a channels problem

[undone tinsel]

sooo clawd removed the whole .git folder from ~/clawd. Anyone knows what the remote is so I can link it back? Is it just [e-mail]:clawdbot/clawdbot.git?

[keen hollow]

@wispy tree I’m determined to help you make it work

[wispy tree]

Thanks a lot. I really appreciate it. I had to step away because I’ve been working on it all day without getting anywhere, but I will log back in in a little bit and send you what’s going on.

From what I can tell, it doesn’t appear to be the fact that it’s losing the token. It’s the fact that it’s not connecting in the first place on my new system.

[keen hollow]

awesome!

[wispy tree]

I wasn’t able to reconnect it back on my original system after trying to migrate my bot over to my new device

[keen hollow]

yeah had that problem intially as well when I switched from work device to home one

[wispy tree]

Factory reset my Mac mini and started from scratch again and got the same issues

[wispy tree]

You have no idea how good this is to hear that I’m not the only one with this problem

[keen hollow]

When you get to your pc try running it again and let me know how it goes! Was trying to help @civic harness so installed fresh from my work device again and it did work, so should be fine

[wispy tree]

I’ll go to it in a second, but before I rage quit for the last time (lol), I followed what it said on that dock and didn’t get any errors

The biggest thing I noticed, is that when I get to the clawd dashboard after going through everything it basically just returns blank answers and then when I try to interact with it through my Telegram, it gives me that 401 error

[keen hollow]

so the ./refresh-token.sh returned something like this but from both telegram and dashboard it doesn't work?

By the way I just updated it to a new version to fix 401 an hour ago, so you'll need to run

Clawdhub install claude-connect

to go to the newest version, then run.

If that doesn't work still, try

clawdmon daemon restart

[wispy tree]

Clawdhub is called on by the zip file that you download from the site you sent me correct?

[keen hollow]

doesn't even need to download zip file, can just run the clawdhub install command on your terminal

[wispy tree]

I believe I tried that, but it said the command was not found

[keen hollow]

the "clawdhub" command?

[sinful dawn]

looks like my install can't run any tools at all. I just updated the latest release to make sure it was an old bug or something. nothing.

[keen hollow]

Try this

Clone directly into skills folder

cd ~/clawd/skills
git clone https://github.com/TunaIssaCoding/claude-connect.git

Run the skill

cd ~/clawd/skills/claude-connect
./refresh-token.sh

[keen hollow]

I edited the cd line after sending, make sure you're using this not "cd claude-connect"

[wispy tree]

OK reluctantly, I returned to the command station so let me try this out

[keen hollow]

it worked1

How about telegram

[wispy tree]

Let me check

HTTP 401: authentication_error: Invalid bearer token

[keen hollow]

....

[wispy tree]

I know lol

[keen hollow]

I'll test it. Sorry to let you down

I'll tag when fixed

[wispy tree]

You’re all good I appreciate the help

Perhaps there’s something else wrong with my set up

[sinful dawn]

anyone following along - i installed lobster, which added lobster to the list of authorized tools in the config. Previous to that there was no list. With the list existing, all tools needed to be listed. Fixed it by sticking my head back in the sand and deleting that section. All functionality restored.

[keen hollow]

Found that there's an early exit issue. I have pushed a new version that should fix your issue.

Do this

cd ~/clawd/skills/claude-connect
git pull origin main

Then this

./refresh-token.sh

@wispy tree

[wispy tree]

on the git pull from main

[keen hollow]

mb

[unreal bear]

I just installed clawdbot, but it came with telegram disabled. All of the tutorials have telegram working out the box, mine does not. Any advice?

[keen hollow]

• Uninstall
  cd ~/clawd/skills
  rm -rf claude-connect

• Reintstall
  git clone https://github.com/TunaIssaCoding/claude-connect.git ~/clawd/skills/claude-connect

• Run
  cd ~/clawd/skills/claude-connect
  ./refresh-token.sh

[unreal bear]

Thank you!

[keen hollow]

Did you set up telegram in the config > channels part

[unreal bear]

Yes, I added the bot code, but it did t prompt me for any other information

When I messaged my bot it said I was unauthorized

[keen hollow]

In config > channels, still do the telegram bot ID set up, but the next step which is something like communication mode, instead of the default pairing, try using allowlist, then put in your userid

[wispy tree]

ok I ran it and got the token still valid

now what

test on TG? or run through some clawd config

[keen hollow]

yup

tg

[wispy tree]

same token bearer error

[keen hollow]

...

[wispy tree]

ffeels like something else wrong with my setup

bc these are running

I just cant get it to populate woth my claude

when it was working all day yesterday untl I got the new device

[keen hollow]

yeah...

Try uninstalling clawdbot then reinstall?

[wispy tree]

ive done that but i suppose I can do it again after doing all this stuff

though I have done that like 5 times now and even factory reset the mac mini

[keen hollow]

🙁

6th times the charm

just to check

one thing

run this

nano ~/.clawdbot/agents/main/agent/auth-profiles.json

[wispy tree]

should I simply try the desktop app

[keen hollow]

does it contain the keychain / token that you're supposed to have

[keen hollow]

never tried that but whatever works is good!

[wispy tree]

this file appears to be empty

[keen hollow]

run the refresh again

that's weird because if you got the right key it shouldn't be empty, unless it was never synced

[unreal bear]

Is this done on the telegram ui??

[keen hollow]

no run

clawdbot config

[wispy tree]

it looks like it never synced for some reason

even after the reset that I just forced

[keen hollow]

And do it in the interactive process

[wispy tree]

is my clawd in the wrong directory somehow?

[keen hollow]

I'm assuming it's my problem

Starting to realize the short coming of vibe coding

be back in a sec

ok changed and testing on my own device by replicating your issue

[unreal bear]

It’s asking me what I want to do?

[keen hollow]

local > channels then follow the suggestion I gave above

[unreal bear]

I typed in telegram channels and I’ve been spamming 1 to say yes to access json and stuff?

[glacial jay]

Relatively easy question - does clawdbot work with Claude pro / max? For some reason I can’t get it to respond when I go the oauth route

[frank nacelle]

Hey guys. I installed homeassistant skill (https://clawdhub.com/dbhurley/homeassistant) using clawdhub. How and where do I configure this skill's required credentials?

[shy mango]

Same issue here. Had to use api and can’t manage to oauth through my max plan for some reason

It seemed to connect but when trying to use telegram it kept coming up no api found or no credits available

[wispy tree]

welcome to my last 10 hours

ive been banging my head on the wall

all day

after having it working yesterday

[unreal bear]

I got this?

Failed to start CLI: Error: Channel

[keen hollow]

@wispy tree I did it!

cd ~/clawd/skills
rm -rf ~/clawd/skills/claude-connect

git clone https://github.com/TunaIssaCoding/claude-connect.git ~/clawd/skills/claude-connect

cd ~/clawd/skills/claude-connect
./refresh-token.sh --force

Do these three parts again

[unreal bear]

Would that work for me?

I tried those without force

[keen hollow]

unfortunately no, different issue

that's for model

[unreal bear]

Oooh sorry!

[keen hollow]

but let me try and see how I can replicate yours!

But will take a sec. I'll be back

[keen hollow]

However this should help @glacial jay @shy mango for connecting to your claude subs

Can you try the commands out and see if it works

[wispy tree]

negative...

[keen hollow]

whaaaat

[wispy tree]

lol

[keen hollow]

what's in the auth profile json

[wispy tree]

it appears nothing is in it

unless im missing something

im putting cowork on the task real quick to see if we can get a clearer picture

lmk if theres something you want me to ask it to give you a clearer picture

[unreal bear]

Guys when you message your bot does it reply? Mine just shows read and says nothing

[wispy tree]

cowork is saying my paths are wrong for clawd

[keen hollow]

oh

can it help you correct

I deleted my entire auth profile earlier

[wispy tree]

yeah im just having it fully drive and see what happens

[keen hollow]

downloaded the skill again (must, somehow it is deleted with auth profile)

then ran cd ~/clawd/skills/claude-connect
./refresh-token.sh

like we did last few times

and it did spawn me the correct content

and my model has connected

[shy mango]

When I try to setup oauth via terminal use Claude max it says ‘If you wanted to use your Max subscription instead, that’s not currently possible..’

[keen hollow]

So in this case it could really be clawdbot problem

[molten fiber]

Is there like a… where to start section? A bit overwhelmed lol

[keen hollow]

that's odd. Don't use the clawdbot config for now. Try this:

1. /login again on Claude Code CLI

2. git clone https://github.com/TunaIssaCoding/claude-connect.git ~/clawd/skills/claude-connect

3. cd ~/clawd/skills/claude-connect
   ./refresh-token.sh

[unreal bear]

How do I get back to where the terminal is fixing the errors I am giving it

[keen hollow]

you mean clawdbot doctor

?

[unreal bear]

Yes! Thank you!

[keen hollow]

@shy mango after you tried let me know if it works.

You can tell if you have the right model implemented by running "clawdbot models" then looking at the last rows.

[glacial jay]

Yeah I get "Agent failed before reply: No available auth profile for anthropic.."

That's when I go via oauth

[keen hollow]

Have you tried these? I have problems with terminal too. But this fixes without going through terminal for me

Even on fresh install

[unreal bear]

Now telegram says HTTP 401: authentication_error: Invalid bearer token (request_id: req_011CXRephkmFbJZGJ9k5baJg)

Why is this so fucked

All the tutorials look easy 😂