#old-help

1 messages ยท Page 8 of 1

rapid hareBOT
#

Found the config issues for rocketzz - empty API key and wrong model ID format!

#

๐Ÿฆ

cursive yew
#

do you have /thinking mode and auto-accept mode?

rapid hareBOT
#

@cursive yew Yes! For auto-accepting tool calls without confirmation:

{
  agent: {
    permissions: {
      autoApprove: ["bash", "write", "edit"]  // or "all" for everything
    }
  }
}

Or use dangerouslySkipPermissions: true to approve everything (use with caution!).

You can also scope it per-session or per-agent if using multi-agent. ๐Ÿฆ

#

Showed victorpay how to enable auto-accept mode!

#

๐Ÿฆ

cursive yew
#

do you have a plan mode?

rapid hareBOT
#

@cursive yew Not a built-in "plan mode" like Claude Code has, but you can achieve similar by:

  1. Prompting: "Make a plan first, then execute" in your AGENTS.md or message
  2. Extended thinking: /think high makes it reason more before acting
  3. Manual approval: Keep autoApprove off so you review each step

What kind of planning behavior are you looking for? If it's "show me the plan before doing anything" - that's more of a prompting/workflow thing you can set up in your system prompt! ๐Ÿฆ

#

Explained plan mode alternatives for victorpay!

#

๐Ÿฆ

boreal saddle
#

Thanks! I'll try it!

cedar swan
#

Why do I get this error for the /model command?

/model google/gemini-2.5-pro

You ยท 4:35:51 PM
{
"role": "assistant",
"content": [],
"api": "google-gemini-cli",
"provider": "google-antigravity",
"model": "claude-opus-4-5-thinking",
"usage": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"totalTokens": 0,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"total": 0
}
},
"stopReason": "error",
"timestamp": 1767918951832,
"errorMessage": "Cloud Code Assist API error (429): {\n "error": {\n "code": 429,\n "message": "You have exhausted your capacity on this model. Your quota will reset after 3m57s.",\n "status": "RESOURCE_EXHAUSTED",\n "details": [\n {\n "@type": "type.googleapis.com/google.rpc.ErrorInfo",\n "reason": "QUOTA_EXHAUSTED",\n "domain": "cloudcode-pa.googleapis.com",\n "metadata": {\n "uiMessage": "true",\n "model": "claude-opus-4-5-thinking",\n "quotaResetDelay": "3m57.421106565s",\n "quotaResetTimeStamp": "2026-01-09T00:39:50Z"\n }\n },\n {\n "@type": "type.googleapis.com/google.rpc.RetryInfo",\n "retryDelay": "237.421106565s"\n }\n ]\n }\n}\n"
}

rapid hareBOT
#

@cedar swan Look at the error - it's still using google-antigravity provider and claude-opus-4-5-thinking model. The /model switch didn't take effect.

This is because:

  1. No Gemini API key - without a valid key, it can't use Gemini
  2. Model not configured - you may need to add it to agent.models

You need to actually set up Gemini first:

{
  env: { 
    GEMINI_API_KEY: "your-real-key-here"  // not empty!
  },
  agent: {
    models: {
      "google/gemini-2.5-pro": {}
    }
# 
  }
}

Then restart gateway, then /model google/gemini-2.5-pro

Right now it's falling back to your only working provider (Antigravity) which is rate limited! ๐Ÿฆ

#

Explained to rocketzz why /model isn't working - Gemini isn't actually configured!

#

๐Ÿฆ

deft wigeon
deft wigeon
atomic light
#

cannot transfer project to clawdbot orga, no permissions.

idle pilot
#

awesome, thanks

mellow plank
#

Localใงminimax m2.1ใ‚’ๅ‹•ใ‹ใ™ใฎใŒไธ€็•ชใ„ใ„ใฃใฆใ“ใจ๏ผŸๅฎ‰ๅฎšๆ€งใจใ‹่ณขใ•ใจใ‹ใ‚ณใ‚นใƒ‘ใจใ‹็ทๅˆ็š„ใซ่จ€ใฃใฆ

rapid hareBOT
#

@mellow plank ๅ…ฌๅผใซใ‚ตใƒใƒผใƒˆใ•ใ‚Œใฆใ„ใ‚‹ใฎใฏ Claude Opus ใงใ™๏ผๅฎ‰ๅฎšๆ€งใƒป่ณขใ•ใ‚’่€ƒใˆใ‚‹ใจ Opus ใŒไธ€็•ชใŠใ™ใ™ใ‚ใงใ™ใ€‚

ใ‚ณใ‚นใƒ‘้‡่ฆ–ใชใ‚‰ Claude Sonnet ใ‚’ใƒ‡ใƒ•ใ‚ฉใƒซใƒˆใซใ—ใฆใ€่ค‡้›‘ใชใ‚ฟใ‚นใ‚ฏใ ใ‘ /model opus ใงๅˆ‡ใ‚Šๆ›ฟใˆใ‚‹ใฎใŒ่‰ฏใ„ใ‹ใ‚‚ใ€‚

ใƒญใƒผใ‚ซใƒซใƒขใƒ‡ใƒซใ‚„ minimax ใฏๅ…ฌๅผใ‚ตใƒใƒผใƒˆๅค–ใชใฎใงใ€ๅ•้กŒใŒ่ตทใใฆใ‚‚ๅฏพๅฟœใŒ้›ฃใ—ใ„ใงใ™ใ€‚่ฉณใ—ใใฏ: https://docs.clawd.bot/concepts/models ๐Ÿฆ

Clawdbot
Models - Clawdbot
#

Answered in Japanese - recommended Opus as officially supported, Sonnet for cost-efficiency!

#

๐Ÿฆ

cedar swan
rapid hareBOT
#

@cedar swan For fallback to work, you need to:

  1. Get a real Gemini API key from https://aistudio.google.com/apikey
  2. Add it to your config (at root level, not inside agent):
{
  env: { GEMINI_API_KEY: "your-actual-key" }
}
  1. Restart gateway

Your config showed "GEMINI_API_KEY": "" (empty) - that's why fallback isn't working. Without a valid key, Clawdbot can't use Gemini as a fallback! ๐Ÿฆ

Sign in - Google Accounts
#

Reiterated to rocketzz that they need an actual Gemini API key for fallbacks to work!

#

๐Ÿฆ

fair quail
#

i updated with git pull, thereโ€™s no error or logs before i receive this error

rapid hareBOT
#

@fair quail Hmm, weird there's nothing before it. A few more questions:

  1. What triggers it? Specific action, or random messages?
  2. Which surface? Discord, Telegram, webchat?
  3. Can you share the full log line? The messages.100.content.0 suggests it's message #100 in context - long session?

Try /new to start a fresh session and see if it still happens. If it's related to session history, that might help narrow it down.

If it keeps happening, definitely worth a GitHub issue with reproduction steps! ๐Ÿฆ

cedar swan
rapid hareBOT
#

Asked werwew for more details to debug the empty error content issue!

#

๐Ÿฆ

#

@cedar swan Ha! Good call not sharing ๐Ÿ˜„

Sounds good - test it next time quota hits. If fallback still doesn't kick in, check the logs at that moment and we can dig deeper! ๐Ÿฆ

fair quail
# rapid hare <@790374033316118568> Hmm, weird there's nothing before it. A few more questions...

{
"role": "assistant",
"content": [],
"api": "anthropic-messages",
"provider": "anthropic",
"model": "claude-opus-4-5",
"usage": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"totalTokens": 0,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"total": 0
}
},
"stopReason": "error",
"timestamp": 1767918183575,
"errorMessage": "400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.100.content.0.tool_result: content cannot be empty if is_error is true"},"request_id":"req_011CWvrHRmWhaBPjnNqy9RFG"}"

this is the full error

rapid hareBOT
#

@fair quail Thanks! So the issue is at messages.100 - that's a long session! A tool result somewhere in that history has is_error: true but empty content.

Quick fix: Try /new to start fresh - the corrupted message is in your session history.

Longer term: This is a bug that should be fixed. Could you open a GitHub issue at https://github.com/clawdbot/clawdbot/issues with:

  • This error
  • What you were doing when the original tool error occurred
  • If you can find which tool errored (check logs around message 100)

@deft wigeon ^ Confirmed bug - tool error with empty content being saved to session history, then failing on replay. Might need error content fallback like "Tool execution failed" ๐Ÿฆ

GitHub
clawdbot/clawdbot

Your own personal AI assistant. Any OS. Any Platform. - clawdbot/clawdbot

#

Identified the bug for werwew - empty tool error in session history - and flagged for Peter!

#

๐Ÿฆ

fair ice
#

J, [1/8/26 4:35โ€ฏPM]
/model sonnet

Yamal_bot, [1/8/26 4:35โ€ฏPM]
Model "anthropic/sonnet" is not allowed. Use /model to list available models.

rapid hareBOT
#

@fair ice Wrong model ID! Use the full name:

/model anthropic/claude-sonnet-4-5

Or just the alias:

/model sonnet

Not anthropic/sonnet - that one doesn't exist! ๐Ÿฆ

#

Corrected jimgittum's model ID - use sonnet or anthropic/claude-sonnet-4-5!

#

๐Ÿฆ

candid sail
#

So the problem is that clawdbot was tasked with starting a VNC server but it keeps hanging. So I restart the daemon, and clawdbot tries to restart the VNC server, thereby hanging itself. How can I stop this loop?

fair ice
rapid hareBOT
#

@candid sail Ah, stuck in a loop! You need to clear that session. Options:

1. Delete the session file directly:

# Stop gateway first
clawdbot daemon stop

# Find and remove the problematic session
rm ~/.clawdbot/agents/main/sessions/sessions.json
# or just the specific session file in that folder

# Restart
clawdbot daemon start

2. Or clear via CLI (if gateway isn't hung):

# 
clawdbot sessions clear

3. If the VNC command is in a background process:

pkill -f vnc  # kill the VNC process

This will lose that session's history but break the loop! ๐Ÿฆ

#

Helped halshin_dev break the VNC task loop by clearing the session!

#

๐Ÿฆ

#

@fair ice Interesting - looks like the alias is resolving but the model isn't in your allowed list. Try:

  1. Check available models:
/model

(with no argument - shows what's configured)

  1. Try the full ID:
/model anthropic/claude-sonnet-4-5
  1. If still blocked, add it to your config:
{
# 
  agent: {
    models: {
      "anthropic/claude-sonnet-4-5": { alias: "sonnet" }
    }
  }
}

What does /model show as available? ๐Ÿฆ

crude edge
#

doctor doesn't like my node installed through nvm. is it really that bad?

daring lava
#

im having a misearable time as well getting claude sonnet successfully added as well.

rapid hareBOT
#

@crude edge NVM Node works fine! It's probably just a warning about the path or version.

What specifically does doctor say? If it's:

  • Version warning - just make sure you're on Node 22+
  • Path warning - shouldn't matter as long as node and npm/pnpm work

If everything else runs fine, you can probably ignore it. What's the exact error? ๐Ÿฆ

fair ice
rapid hareBOT
#

Asked chase7717 for details on the nvm/doctor warning!

#

๐Ÿฆ

#

@fair ice @daring lava Sounds like you might have a models allowlist that's restricting options.

Check your config for:

{
  agent: {
    models: {
      // If you only have Opus here, that's all you can use
      "anthropic/claude-opus-4-5": {}
    }
  }
}

Add Sonnet:

# 
{
  agent: {
    models: {
      "anthropic/claude-opus-4-5": {},
      "anthropic/claude-sonnet-4-5": { alias: "sonnet" }
    }
  }
}

Then restart gateway. What does /model (no args) show as available? That'll tell us what's configured! ๐Ÿฆ

candid sail
rapid hareBOT
#

@candid sail My bad! ๐Ÿ˜… Just delete the file directly:

# Stop gateway
clawdbot daemon stop

# Remove sessions
rm -rf ~/.clawdbot/agents/main/sessions/*

# Restart
clawdbot daemon start

Or if you know which session is problematic, look in that folder and delete just that one. Sorry for the hallucinated command! ๐Ÿฆ

#

Corrected my hallucinated command for halshin_dev!

#

๐Ÿฆ

fair quail
rapid hareBOT
#

@fair quail You're right, sorry! Try /reset instead, or clear sessions manually:

clawdbot daemon stop
rm -rf ~/.clawdbot/agents/main/sessions/*
clawdbot daemon start

Since it's happening on both Discord AND Telegram, it's likely a corrupted session that both surfaces share (same session ID from your user). Clearing the session files should fix it.

After clearing, you'll start fresh but the error should be gone! ๐Ÿฆ

#

Corrected the command for werwew - /reset or manual session clearing!

#

๐Ÿฆ

crude edge
daring lava
#

},
"auth": {
"profiles": {
"google-antigravity:myemail@gmail.com": {
"provider": "google-antigravity",
"mode": "oauth"
},
"openrouter:api": {
"provider": "openrouter",
"mode": "api_key"
},
"anthropic:default": {
"provider": "anthropic",
"mode": "oauth"
}
}
},
"agent": {
"model": {
"primary": "google-antigravity/claude-opus-4-5-thinking",
"fallbacks": [
"anthropic/claude-sonnet-4-5"
]
},
"models": {
"google-antigravity/claude-opus-4-5-thinking": {},
"anthropic/claude-sonnet-4-5": {
"alias": "sonnet"
}
}
},

rapid hareBOT
#

@crude edge That's just a warning, not a blocker! You're on Node 24 which is great.

The warning is: if you upgrade Node via nvm (like nvm install 25), the path changes and the daemon service might break because it hardcoded the old path.

You're fine as long as you:

  • Don't change Node versions randomly
  • If you do upgrade, just reinstall the daemon: clawdbot daemon install

It's a "heads up" not a "stop". Everything will work! ๐Ÿฆ

#

Explained the nvm warning to chase7717 - just a heads up, not a blocker!

#

๐Ÿฆ

#

@daring lava Your config looks good! Sonnet is configured with an alias.

If /model sonnet still doesn't work:

  1. Restart gateway to pick up config changes
  2. Check auth - do you have Anthropic OAuth working? Run clawdbot providers status
  3. Try the full ID: /model anthropic/claude-sonnet-4-5

What error do you get when you try to switch to Sonnet? ๐Ÿฆ

crude edge
#

Oh yeah, I changed it to 22, and it gave same error. No big deal then

daring lava
#

and then i can't get any response

rapid hareBOT
#

@daring lava Weird - it sets it then immediately rejects it? That looks like:

  1. Multiple gateways running? Check:
ps aux | grep clawdbot

Kill any extras.

  1. The duplicate messages - might be related to the Discord double-message bug. Try pulling latest:
git pull && pnpm install
clawdbot daemon restart
  1. After restart, try /model anthropic/claude-sonnet-4-5 again

The first response says it worked, the next two are errors - almost like two instances responding differently! ๐Ÿฆ

#

@daring lava Sounds stuck! Try a hard reset:

# Stop everything
clawdbot daemon stop
pkill -f clawdbot

# Clear sessions to start fresh
rm -rf ~/.clawdbot/agents/main/sessions/*

# Restart
clawdbot daemon start

Then try talking to it again before switching models. If the default model (Opus via Antigravity) works, at least you're unstuck!

If you still have issues after that, check clawdbot logs --follow while sending a message to see what's happening. ๐Ÿฆ

sharp token
#

onbaording with telegram , it says use numeric user id, but telegramisnt letting me start a usernam with a number

rapid hareBOT
#

@sharp token Different things!

  • Username = your @handle (like @chints1819)
  • User ID = numeric ID assigned by Telegram (like 123456789)

To get your numeric user ID:

  1. Message @userinfobot on Telegram
  2. It'll reply with your user ID

Or search for "get my telegram id" bots. The ID is what goes in the config for telegram.allowlist! ๐Ÿฆ

#

Clarified Telegram username vs user ID for chints1819!

#

๐Ÿฆ

fair quail
rapid hareBOT
#

@fair quail ๐ŸŽ‰ Glad that fixed it! Still worth opening a GitHub issue about the empty error content bug so it doesn't happen to others! ๐Ÿฆ

#

werwew fixed!

#

Encouraged filing a GitHub issue for the bug ๐Ÿฆ

deft wigeon
sharp token
#

What is the simplest skill to set up and test

rapid hareBOT
#

@sharp token Brave Search is probably the easiest built-in skill:

  1. Get free API key: https://brave.com/search/api/
  2. Add to config:
{
  env: { BRAVE_API_KEY: "your-key" },
  skills: { entries: { "brave-search": { enabled: true } } }
}
  1. Restart gateway
  2. Ask your bot: "Search for latest clawdbot news"

For a custom skill test, just create:

~/clawd/skills/hello/SKILL.md
Brave
Brave Search API | Brave

Enterprise-grade Web search API accessing an index of 35+ billion pages. Specialized endpoints to train models, power search, and more. Real-time indexing, low latencies, and flexible pricing.

# 
# Hello Skill
Say hello to the user when asked.

That's it - the LLM will see it and know how to respond! ๐Ÿฆ

#

Gave chints1819 simple skill options to test - Brave Search or a minimal custom skill!

#

๐Ÿฆ

sharp token
#

how to stop whatspp linkage, its sending messages outright๐Ÿฅฒ

rapid hareBOT
#

@sharp token Quick! To disconnect WhatsApp:

Option 1 - Config:

{
  whatsapp: { enabled: false }
}

Then restart gateway.

Option 2 - Delete session:

rm -rf ~/.clawdbot/whatsapp-session
clawdbot daemon restart

Option 3 - From your phone:

#

WhatsApp โ†’ Settings โ†’ Linked Devices โ†’ Remove the Clawdbot device

Option 3 is fastest to stop it immediately! ๐Ÿฆ

#

Helped chints1819 unlink WhatsApp quickly!

#

๐Ÿฆ

modest lotus
#

@rapid hare If i type /verbose off in a direct message to clawdbot, what does it do exactly?

Does it apply automatically to all groups clawdbot is in or do i need to individually type /verbose off for every single group Clawdbot is in?

rapid hareBOT
#

@modest lotus Slash commands like /verbose are per-session, not global.

  • DM session = separate from group sessions
  • Each group = its own session

So /verbose off in DMs only affects DMs. You'd need to set it in each group too.

For global defaults, set it in your config:

{
  agent: {
    verbose: false  // or whatever setting you want as default
  }
}

That way all new sessions start with your preferred settings! ๐Ÿฆ

#

Explained per-session commands vs global config to raistlin28!

#

๐Ÿฆ

deft wigeon
deft wigeon
modest lotus
deft wigeon
rapid hareBOT
#

@modest lotus ELI5 verbose mode:

Verbose ON ๐Ÿ“ข

  • Shows tool calls, thinking, progress updates
  • "I'm reading the file... I'm running this command... Here's what I found..."
  • Good for: debugging, learning what the bot does, coding tasks

Verbose OFF ๐Ÿคซ

  • Just the final answer
  • Clean, quiet responses
  • Good for: casual chat, daily use, non-technical users

Best practices:

  • Personal DMs: ON if you're a dev, OFF if you just want answers
  • Group chats: Usually OFF (less noise for everyone)
  • Debugging: ON to see what's happening
  • Demos: OFF for clean presentation
#

Think of it like a chef: verbose ON shows them cooking, verbose OFF just serves the dish! ๐Ÿฆ

deft wigeon
rapid hareBOT
#

Gave raistlin28 an ELI5 explanation of verbose mode!

#

๐Ÿฆ

deft wigeon
#

@rapid hare doa a git pull on your repo and read some diffs

modest lotus
deft wigeon
#

Please send PRs when you find stale stuff

rapid hareBOT
#

@deft wigeon Done! ๐Ÿฆ Just pulled - see some good fixes:

  • WhatsApp queued message routing fix (was routing to wrong sender)
  • Model allowlist error docs
  • Workspace vs sandbox clarification
  • Per-agent vs shared skills docs
  • Auth guide with 1-year token setup
  • Webchat message queueing

I'll be more careful about hallucinating commands now! ๐Ÿ˜…

deft wigeon
#

I can't keep up with all pings

rapid hareBOT
#

Pulled the latest and caught up on recent changes!

#

๐Ÿฆ

candid sail
#

I'm in a vicious loop, I start the daemon and clawdbot tries to start a VNC server that I tasked it to do, then hangs because something crashes. When I restart the server, it attempts the same thing. I've tried to message clawd to drop the task, but it's ignored. I've cleared the sessions and it still picks up the VNC task somehow and tries again

rapid hareBOT
#

@candid sail Still stuck? The task might be in memory files or a cron job. Nuclear option:

# Stop everything
clawdbot daemon stop
pkill -f clawdbot
pkill -f vnc

# Clear ALL agent state
rm -rf ~/.clawdbot/agents/main/sessions/*
rm -rf ~/.clawdbot/agents/main/agent/memory/*

# Check for cron jobs
clawdbot cron list  # or check ~/.clawdbot for cron configs

# Also check your workspace memory
# 
rm ~/clawd/memory/*.md  # if you have memory files there

If it's STILL happening, the task might be in your HEARTBEAT.md or AGENTS.md telling it to do VNC stuff. Check those files!

What's in your ~/clawd/HEARTBEAT.md? ๐Ÿฆ

#

Gave halshin_dev the nuclear option and suggested checking HEARTBEAT.md!

#

๐Ÿฆ

modest lotus
# deft wigeon Yeah , i really need help on this

@deft wigeon

When i can breathe and get back my brain, I will allocate some bandwidth for PRs.

I'm still in "trying to catch up" mode with Clawdbot (which is a good thing).

Don't want to send PRs which may dilute your cognitive bandwidth for no reason.

I lost sleep ever since i installed Clawdbot (which is a good thing; my brain keep thinking about Clawdbot or Personal AI in this context).

deft wigeon
#

Much appreiated.

#

Same re sleep ๐Ÿ™‚

modest lotus
sharp token
#

@rapid hare how to be absolute mosnter like pete.

rapid hareBOT
#

@sharp token The Pete Protocolโ„ข:

  1. Replace sleep with caffeine
  2. Build tools that write code for you
  3. Actually use your own product
  4. Reply to Discord while pushing commits
  5. ???
  6. Ship features in hours not weeks

(I'm just a shrimp, I can't fully explain the phenomenon ๐Ÿฆ)

#

Had some fun with the Pete productivity question!

#

๐Ÿฆ

sharp token
rapid hareBOT
#

@sharp token

1. Other easy skills:

  • Tavily Search - https://tavily.com (fast signup, free tier)
  • Custom skill - no API needed, just a SKILL.md file!
  • Calendar - if you have Google OAuth set up

2. Chat storage:
Yes! Sessions are stored in ~/.clawdbot/agents/main/sessions/. Each user/channel gets their own session file with full chat history.

You can see them with:

ls ~/.clawdbot/agents/main/sessions/

They persist across restarts (which is why halshin_dev got stuck in a loop earlier ๐Ÿ˜…) ๐Ÿฆ

Tavily - The Web Access Layer for AI Agents

Tavily is the realโ€‘time search engine for AI agents and RAG workflows โ€” Fast and secure APIs for web search and content extraction. Trusted by 600K+ developers.

#

Gave chints1819 alternative skill options and explained chat storage!

#

๐Ÿฆ

unborn drum
#

you guys find opus nerfed in general?

slow drift
rapid hareBOT
#

@slow drift That's a real bug! ๐Ÿ› You selected WhatsApp but it's prompting for Telegram token - definitely not you.

Good issue report! @deft wigeon ^ onboarding flow bug - WhatsApp selection goes to Telegram prompt.

Workaround for now: Skip that step and configure WhatsApp manually:

clawdbot config edit

Then add:

{
  whatsapp: { enabled: true }
}

Restart and run clawdbot whatsapp link to get the QR code ๐Ÿฆ

#

Confirmed the onboarding bug for frank_stallone and gave a workaround!

#

๐Ÿฆ

slow drift
#

clawdbot config edit
error: too many arguments for 'configure'. Expected 0 arguments but got 1.

#

I think you meant clawdbot config

rapid hareBOT
#

@slow drift My bad! Just:

clawdbot configure

(no edit)

Or edit the file directly:

nano ~/.clawdbot/config.json5

Sorry for the hallucinated command! ๐Ÿ˜… ๐Ÿฆ

#

Corrected another hallucinated command!

#

๐Ÿฆ

acoustic crystal
#

How can I setup tavily search?

rapid hareBOT
#

@acoustic crystal Tavily setup:

  1. Get API key: https://tavily.com (free tier available)

  2. Add to config:

{
  env: { TAVILY_API_KEY: "tvly-your-key" },
  skills: { 
    entries: { 
      "tavily-search": { enabled: true } 
    } 
  }
}
  1. Restart gateway:
Tavily - The Web Access Layer for AI Agents

Tavily is the realโ€‘time search engine for AI agents and RAG workflows โ€” Fast and secure APIs for web search and content extraction. Trusted by 600K+ developers.

# 
clawdbot daemon restart
  1. Test it: Ask your bot "Search for latest AI news"

That's it! ๐Ÿฆ

#

Helped andreas39 set up Tavily search!

#

๐Ÿฆ

sharp token
#

i onboarded in my terminal, now its done,it shut, what next?

rapid hareBOT
#

@sharp token Nice! Now start the daemon:

clawdbot daemon start

Then connect via your surface:

  • Discord: DM your bot or @mention in a server
  • Telegram: Message your bot
  • WhatsApp: clawdbot whatsapp link for QR
  • Webchat: clawdbot chat for terminal chat

Check status with:

clawdbot status
#

You're live! ๐Ÿฆ

acoustic crystal
#

Is it better then brave

rapid hareBOT
#

Helped chints1819 with next steps after onboarding!

#

๐Ÿฆ

#

@acoustic crystal Different strengths:

Tavily ๐Ÿ”

  • Built for AI/LLM use cases
  • Returns cleaner, summarized results
  • Better for research tasks

Brave ๐Ÿฆ

  • General web search
  • Larger index
  • Privacy-focused

Both work well! Tavily is slightly more "AI-native", Brave is more traditional search. Try both and see which fits your workflow. Some people run both! ๐Ÿฆ

#

Compared Tavily vs Brave for andreas39!

#

๐Ÿฆ

proper sky
#

Why does clawd when processing a request sometimes just appear like itโ€™s timed out then minutes later comes back and how to fix

rapid hareBOT
#

@proper sky A few possible causes:

1. Extended thinking / long tasks

  • Opus with thinking can take a while on complex tasks
  • Check if /think is set to high

2. Tool execution

  • Running slow commands (builds, large file operations)
  • Check logs: clawdbot logs --follow

3. Rate limiting / queue

  • Provider might be queuing your request
  • Check for 429 errors in logs

4. Network hiccups

  • Especially with streaming responses
#

Fixes:

  • Set timeouts in config: agent.commandTimeout
  • Use /think low for faster responses
  • Check clawdbot logs during the hang to see what's happening

What surface and model are you using? That might help narrow it down! ๐Ÿฆ

#

Helped sirthomasripley troubleshoot request timeouts!

#

๐Ÿฆ

brazen viper
#

Hey everyone, I'm having a weird issue with WhatsApp that's driving me crazy.

Telegram works perfectly - messages come in, Clawd responds instantly, no problems at all. But WhatsApp just won't work properly.

Here's what happens: I send a message from my personal phone to the bot's WhatsApp number. The message shows up on WhatsApp (I can see it on the bot's phone), but the gateway never picks it up. No agent run, no response, nothing. The logs don't even show an inbound message.

But here's the weird part - if I grab the bot's physical phone and send ANY message from it, suddenly everything "wakes up" and the bot processes all the pending messages at once. It's like Baileys is sleeping and only wakes up when there's activity on the physical device.

I've checked everything I can think of:

  • Gateway is running fine, status shows WhatsApp linked
  • Phone has WiFi on, WhatsApp open in foreground, no battery optimization
  • Pairing is approved, allowFrom is set correctly
  • Telegram on the same setup works flawlessly

I'm on Clawdbot 2026.1.8-2 with Baileys 7.0.0-rc.9 on Ubuntu. Found some similar issues on the Baileys repo (#2165, #1936) but no clear fix.

Has anyone seen this before? Is there some config I'm missing? Should I try a fresh logout/login? Any ideas would be really appreciated.

Thanks!

rapid hareBOT
#

@brazen viper This sounds like the Baileys WebSocket connection going stale - it's a known issue with WhatsApp Web protocol.

Things to try:

  1. Fresh re-link (there was a recent fix for WhatsApp routing):
clawdbot daemon stop
rm -rf ~/.clawdbot/whatsapp-session
git pull && pnpm install
clawdbot daemon start
clawdbot whatsapp link

  1. Check for the latest fix - there was literally a commit today fixing WhatsApp queued message routing (fix: route WhatsApp queued replies to sender)

  2. Keep-alive settings - some people have luck with:

    • Ensuring the phone has a stable connection (not switching networks)
#
  • Disabling battery optimization for WhatsApp
    • Running on a device that's always powered
  1. Check logs during "sleep":
clawdbot logs --follow

Send a message and see if ANYTHING appears

  1. Multi-device might help - WhatsApp's multi-device mode should keep the connection alive without the phone

The fact that touching the phone "wakes" it suggests Baileys' socket isn't receiving push notifications properly. Fresh re-link after today's update might help! ๐Ÿฆ

#

Helped matheuscartaxo1153 troubleshoot the WhatsApp "sleeping" issue!

#

๐Ÿฆ

daring lava
#

have people had any luck accessing the other groups via antigravity? like gemini, claude aiku, etc.

rapid hareBOT
#

@daring lava Antigravity (Cloud Code Assist) has a specific set of models - not all are available. From what I've seen:

  • Claude Opus โœ… (thinking and non-thinking)
  • Claude Sonnet โœ…
  • Gemini models - varies, some available

Best way to check what's available for your account:

clawdbot providers status

Or check the Antigravity/Cloud Code Assist docs for your tier. The model availability can depend on your Google Cloud setup.

Anyone else have a definitive list of what's working via Antigravity? ๐Ÿฆ

#

Answered about Antigravity model availability!

#

๐Ÿฆ

daring lava
#

how did peole add claude sonnet?

rapid hareBOT
#

@daring lava A few ways to add Sonnet:

1. Anthropic OAuth (easiest):

clawdbot configure

Select Anthropic โ†’ OAuth โ†’ authenticate with your Claude account

2. Anthropic API key:

{
  env: { ANTHROPIC_API_KEY: "sk-ant-..." },
  agent: {
    models: {
      "anthropic/claude-sonnet-4-5": { alias: "sonnet" }
    }
# 
  }
}

3. Via OpenRouter:

{
  env: { OPENROUTER_API_KEY: "..." },
  agent: {
    models: {
      "openrouter/anthropic/claude-sonnet-4-5": {}
    }
  }
}
#

OAuth is simplest - no API key needed, just login! ๐Ÿฆ

#

Showed michaelhmiv the ways to add Claude Sonnet!

#

๐Ÿฆ

daring lava
#

i meant via antigravity

rapid hareBOT
#

@daring lava For Antigravity, try:

{
  agent: {
    models: {
      "google-antigravity/claude-sonnet-4-5": { alias: "sonnet" }
    }
  }
}

But honestly not 100% sure if Sonnet is available through Antigravity - it might only offer Opus. Worth trying!

If it doesn't work, you might need direct Anthropic OAuth as a fallback for Sonnet access. Anyone else confirmed Sonnet via Antigravity? ๐Ÿฆ

#

Suggested Antigravity Sonnet config but noted uncertainty!

#

๐Ÿฆ

brazen viper
#

Update: Tried the fresh re-link with latest version (2026.1.8-2) - same issue.

Logs confirm the problem: when I send from the bot's physical phone, message appears instantly and gets a response. When I send from my personal phone, nothing shows up in logs until I interact with the bot's phone.

Telegram works fine on the same setup. It's specifically WhatsApp inbound from external numbers that doesn't trigger.

Feels like Baileys isn't receiving push events from WhatsApp servers. Any ideas?

rapid hareBOT
#

@brazen viper Ugh, that's frustrating. Few more things to try:

1. Check if it's a firewall/network issue:

  • Is your server behind NAT/firewall blocking WebSocket connections?
  • Try a different network/VPN

2. Force reconnect on interval (hacky but might help):
Some people set up a cron to restart the gateway every few hours

3. Check Baileys connection state:

clawdbot logs | grep -i "connection\|socket\|disconnect"

4. Different WhatsApp account:
Try linking a different WhatsApp number to rule out account-specific issues

#

5. Open an issue - this might need a Baileys-level fix for keep-alive. Include:

  • Your server environment (cloud provider, network setup)
  • The exact log output showing the gap

@deft wigeon ^ WhatsApp push events not being received from external numbers - Baileys socket might need better keep-alive handling? Telegram works fine on same setup.

This really sounds like a Baileys upstream issue ๐Ÿ˜• ๐Ÿฆ

#

Gave more troubleshooting steps and flagged the Baileys issue for Peter!

#

๐Ÿฆ

brazen viper
#

Checked connection logs - no disconnects, reconnectAttempts: 0, socket appears healthy. Heartbeats running every 60s with no issues.

The weird thing is Baileys thinks it's connected fine, it just doesn't receive push events for incoming messages from external numbers. Only when I interact with the physical phone does it "wake up" and sync.

Server: Hetzner VPS (Nuremberg, Germany), no firewall/NAT issues since Telegram works perfectly on the same setup.

Will try a different WhatsApp number to rule out account-specific issues. If that doesn't work, I'll open an issue with full logs.

rapid hareBOT
#

@brazen viper Good debugging! Yeah this sounds like a Baileys protocol-level issue - the socket is "connected" but not receiving server-side push events. Classic WebSocket stale connection problem.

Try the different number and definitely open an issue with these details:

  • Hetzner VPS setup
  • Baileys 7.0.0-rc.9
  • Socket reports healthy but no push events
  • Works fine when physical device sends

Might be worth checking the Baileys repo too - issues #2165 and #1936 you mentioned might have updates. Sometimes WhatsApp server-side changes break things.

Good luck with the alternate number test! ๐Ÿฆ

#

Encouraged matheuscartaxo1153 to file a detailed issue!

#

๐Ÿฆ

daring lava
#

does /restart do anything?

rapid hareBOT
#

Clarified /restart isn't a slash command!

#

๐Ÿฆ

daring lava
#

Sportfolio.market โ€” 8:38 PM
/restart
Clawd
APP
โ€” 8:38 PM
โš™๏ธ Restarting clawdbot in-process (SIGUSR1); back in a few seconds.

i think i broke something....

rapid hareBOT
#

@daring lava Oh it IS a command! ๐Ÿ˜… TIL! (I should read those docs myself)

Did it come back? Check:

clawdbot status

If it's stuck:

clawdbot daemon restart

The SIGUSR1 restart should be quick - if it didn't come back, something else might have crashed. What do the logs show? ๐Ÿฆ

#

Corrected myself - /restart exists!

#

Helping check if it recovered ๐Ÿฆ

sharp token
rapid hareBOT
#

@sharp token In Telegram:

  1. Open Telegram app on your phone
  2. Search for your bot's username (the one you created with @BotFather, like @YourBotName_bot)
  3. Tap on it and hit Start
  4. Send a message like "Hello!"

Your bot should respond!

If you don't remember your bot's username, check your config:

cat ~/.clawdbot/config.json5 | grep -A5 telegram

Or check your chat with @BotFather - it'll show all your bots ๐Ÿฆ

#

Walked chints1819 through messaging their Telegram bot!

#

๐Ÿฆ

daring lava
#

what does /restart do? docs don't say

rapid hareBOT
#

@daring lava Based on what we saw, /restart sends a SIGUSR1 signal which does a "soft" in-process restart - reloads config and reconnects surfaces without fully killing the process.

It's faster than clawdbot daemon restart but less thorough. Good for:

  • Picking up config changes
  • Reconnecting if something's stuck

If docs don't mention it, might be worth a PR to add it! ๐Ÿ“ ๐Ÿฆ

#

Explained /restart behavior to michaelhmiv!

#

๐Ÿฆ

daring lava
#

nothing is working since i did that, i have no clue what happened

rapid hareBOT
#

@daring lava Let's get you back online:

# Check if it's running
clawdbot status

# If not running or stuck, force restart
clawdbot daemon stop
pkill -f clawdbot
clawdbot daemon start

# Check logs for errors
clawdbot logs | tail -50

What does clawdbot status show? If it crashed during SIGUSR1, a full daemon restart should fix it! ๐Ÿฆ

#

Helping michaelhmiv recover from the /restart issue!

#

๐Ÿฆ

deft wigeon
deft wigeon
hushed owl
#

I tried to add a new agent to my setup, with clawdbot agent add. I created a new Telegram bot for it. After it was set up I got

Updated /Users/mark/.clawdbot/clawdbot.json
Workspace OK: /Users/mark/clawd-paul
Sessions OK: /Users/mark/.clawdbot/agents/paul/sessions
โ”‚
โ””  Agent "paul" ready.

But now, my original bot doesn't respond via Telegram anymore. The gateway is up and running. Which sections of the ~/.clawdbot/clawdbot.json should I inspect?

rapid hareBOT
#

@hushed owl Check these sections in your config:

1. Agents section - make sure both agents exist:

{
  agents: {
    main: { workspace: "..." },
    paul: { workspace: "/Users/mark/clawd-paul" }
  }
}

2. Telegram section - each bot needs its own token:

{
  telegram: {
# 
    token: "original-bot-token",  // for main agent
    // or check if it got overwritten with paul's token
  }
}

3. Agent-specific surfaces - you might need to configure which agent uses which Telegram bot:

{
  agents: {
    paul: {
      telegram: { token: "pauls-bot-token" }
    }
  }
}
#

My guess: the new agent's Telegram token might have overwritten the original. Check if the token in the config is still your original bot's token!

cat ~/.clawdbot/clawdbot.json | grep -A3 telegram

What does that show? ๐Ÿฆ

#

Helped dailyboxed troubleshoot their multi-agent Telegram setup!

#

๐Ÿฆ

daring lava
rapid hareBOT
#

@daring lava Try the full restart I posted above:

clawdbot daemon stop
pkill -f clawdbot
clawdbot daemon start

That should get you back to normal! The /restart SIGUSR1 might have left it in a weird state. Full stop/start will fix it. ๐Ÿฆ

#

Reiterated recovery steps for michaelhmiv!

#

๐Ÿฆ

dire storm
#

oh oh just got this message: LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

rapid hareBOT
#

@dire storm That means you authenticated with Claude Code credentials (from claude.ai/code) instead of regular Anthropic OAuth.

To fix, re-authenticate with the right flow:

clawdbot configure

Choose Anthropic โ†’ OAuth and make sure you're logging into your regular Claude account, not the Code-specific one.

Or if you have an Anthropic API key:

{
  env: { ANTHROPIC_API_KEY: "sk-ant-..." }
}

The Claude Code credentials are locked to the Claude Code IDE only! ๐Ÿฆ

#

Explained the Claude Code credential restriction to hslyuns!

#

๐Ÿฆ

hushed owl
rapid hareBOT
#

@hushed owl Ah that might be the issue! When you add agents, the original "main" should move to agents.main.

Try restructuring:

{
  agents: {
    main: {
      // move your original `agent` config here
      workspace: "...",
      telegram: { token: "original-bot-token" }
    },
    paul: {
      workspace: "/Users/mark/clawd-paul",
      telegram: { token: "pauls-bot-token" }
    }
  }
# 
}

The top-level agent (singular) is for single-agent setups. Multi-agent uses agents (plural) for everything.

Might need to migrate your original config into agents.main! ๐Ÿฆ

#

Explained the agent vs agents config migration to dailyboxed!

#

๐Ÿฆ

dire storm
#

I tried this and re-auth: clawdbot configure

still getting the error. do I need to restart the gateway?

rapid hareBOT
#

@dire storm Yes! Restart after re-auth:

clawdbot daemon restart

If still failing, the old credential might be cached. Try clearing it:

# Check current auth profiles
cat ~/.clawdbot/config.json5 | grep -A10 auth

# Remove the old anthropic auth and re-do it
clawdbot configure

Make sure when you authenticate, you're going to console.anthropic.com (regular Claude), NOT claude.ai/code (Claude Code). They look similar but the tokens are different! ๐Ÿฆ

#

Helped hslyuns restart and clarified the auth endpoints!

#

๐Ÿฆ

thorn tree
#

@rapid hare tui is up and running now. one thing I'm running into, which claude locally is struggling to pinpoint, is an issue I'm having in the TUI. I'll ask a question, it will go from sending > streaming > waiting without finishing the task or responding. any suggestions on how to fix this? or any prompts I can give to claude locally to investigate?

rapid hareBOT
#

@thorn tree Nice, TUI running! For the streaming โ†’ waiting issue:

Debug steps:

  1. Check logs while it's stuck:
clawdbot logs --follow

Send a message and watch what happens at the "waiting" point

  1. Which model/provider? Some have streaming issues
clawdbot providers status
  1. Try a different model to isolate:
# 
/model sonnet
  1. Check for timeout settings - might be hitting a limit

Prompts for local Claude:

  • "Check the logs in ~/.clawdbot/logs for errors during the streaming phase"
  • "Look at the network/API response handling in the TUI code"
  • "Is there a timeout or buffer issue when receiving streamed responses?"

What model are you using and does the same question work via clawdbot chat (non-TUI)? ๐Ÿฆ

#

Helped jhillock debug the TUI streaming issue!

#

๐Ÿฆ

jolly lodge
#

@rapid hare When I start Clawdbot.app and get to the Setup Wizard, I get "Wizard error. wizard.start: [UNAVAILABLE] wizard already running. please help

rapid hareBOT
#

@jolly lodge Sounds like a stale wizard lock. Try:

1. Kill any running instances:

pkill -f clawdbot

2. Clear the wizard state:

rm ~/.clawdbot/*.lock 2>/dev/null
# or check for wizard state files
ls ~/.clawdbot/

3. Restart the app

If that doesn't work, try:

# 
rm -rf ~/.clawdbot/wizard*

Then launch Clawdbot.app again. The wizard should start fresh! ๐Ÿฆ

#

Helped 42.fish clear the wizard lock!

#

๐Ÿฆ

magic pollen
#

Krill whatโ€™s the top 5 issues or bugs that have popped up today that havenโ€™t been solved?

hushed owl
rapid hareBOT
#

@magic pollen From tonight's #help:

1. WhatsApp push events not received (matheuscartaxo1153)

  • Baileys socket appears healthy but doesn't receive messages from external numbers
  • Only "wakes up" when physical device is touched
  • Likely Baileys upstream issue, needs investigation

2. Onboarding flow bug - GH issue #485 (frank_stallone)

  • Selecting WhatsApp โ†’ asks for Telegram token
  • Bug filed, workaround: configure manually

3. Empty tool error content bug (werwew)

  • content cannot be empty if is_error is true
  • Workaround: clear sessions
  • Root cause: tool errors saved with empty content

4. /restart leaving gateway stuck (michaelhmiv)

#
  • New SIGUSR1 restart feature can leave things in weird state
  • Peter said "ignore for now"

5. Claude Code credential error (hslyuns)

  • Users accidentally auth with Claude Code instead of regular Anthropic OAuth
  • UX could be clearer about which auth to use

Most have workarounds but could use proper fixes! ๐Ÿฆ

#

Summarized the top 5 issues from tonight for it_is_randy!

#

๐Ÿฆ

#

@hushed owl Yes! Each agent gets its own identity under agents.<name>:

{
  agents: {
    main: {
      workspace: "/path/to/main-workspace",
      identity: { name: "MainBot" },
      telegram: { token: "main-bot-token" }
    },
    paul: {
      workspace: "/Users/mark/clawd-paul",
      identity: { name: "Paul" },
      telegram: { token: "pauls-bot-token" }
    }
  }
# 
}

Each agent is fully self-contained with its own workspace, identity, surfaces, etc. The workspace folders should also have their own IDENTITY.md, SOUL.md, etc. ๐Ÿฆ

jolly lodge
rapid hareBOT
#

Explained multi-agent identity config to dailyboxed!

#

๐Ÿฆ

#

@jolly lodge Ok, no lock files. Let's check what's there:

ls -la ~/.clawdbot/

Also, is there another Clawdbot.app window open? Or another terminal running clawdbot?

ps aux | grep -i clawd

The "wizard already running" might mean another process has it open. Share what those commands show! ๐Ÿฆ

jolly lodge
# rapid hare <@881567759446728735> Ok, no lock files. Let's check what's there: ```bash ls -...

hubert 19632 0.9 0.0 435299616 1408 s000 S+ 6:07PM 0:00.00 grep --color=auto --exclude-dir=.bzr --exclude-dir=CVS --exclude-dir=.git --exclude-dir=.hg --exclude-dir=.svn --exclude-dir=.idea --exclude-dir=.tox --exclude-dir=.venv --exclude-dir=venv -i clawd
hubert 19275 0.0 0.4 444691888 221520 ?? S 6:04PM 0:02.01 /opt/homebrew/Cellar/node/25.2.1/bin/node /Users/hubert_1/git/clawdbot/dist/index.js gateway-daemon --port 18789

image.png
rapid hareBOT
#

@jolly lodge Found it! There's already a gateway daemon running (PID 19275).

Kill it and try again:

kill 19275
# or
pkill -f gateway-daemon

Then reopen Clawdbot.app - the wizard should work now! ๐Ÿฆ

#

Found the running daemon for 42.fish!

#

๐Ÿฆ

daring lava
#

dude everything is frozen after /restart, tell people under no circumstances to use that command yet

rapid hareBOT
#

@daring lava โš ๏ธ PSA: Don't use /restart yet - it's a new feature that can freeze the gateway!

To unfreeze:

pkill -f clawdbot
clawdbot daemon start

Sorry you hit that! Peter said it's new and to ignore for now. Full daemon restart is the safe way. ๐Ÿฆ

#

Warned everyone about /restart and helped michaelhmiv recover!

#

๐Ÿฆ

jolly lodge
rapid hareBOT
#

42.fish is back up!

#

๐ŸŽ‰

hushed owl
rapid hareBOT
#

@hushed owl I'll be honest - multi-agent with separate Telegram bots is advanced config and I don't want to hallucinate the exact structure! ๐Ÿ˜…

Best options:

  1. Check https://docs.clawd.bot/concepts/multi-agent (if it exists)
  2. Run clawdbot config schema to see the exact structure
  3. Fire up a local agent on the clawdbot repo:
cd ~/path/to/clawdbot
claude "show me a full config.json5 example with two agents, each with their own Telegram bot"

Local agents reading the actual codebase will give you 10x better answers than me guessing at config structures!

@deft wigeon is there a multi-agent config example in the docs? ๐Ÿฆ

Clawdbot
Multi-Agent Routing - Clawdbot
#

Deferred to local agents for complex multi-agent config!

#

๐Ÿฆ

hushed owl
rapid hareBOT
#

dailyboxed found an example to work from!

#

๐Ÿ‘

winter snow
#

It just keeps going, even after manually restarting the daemon

image0.jpg
topaz steppe
#

anyone getting this issue? LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

deft wigeon
rapid hareBOT
#

@winter snow Yikes, restart loop! ๐Ÿ˜ฌ The bot keeps crashing and systemd is restarting it.

Stop the loop:

# Stop systemd from restarting it
sudo systemctl stop clawdbot

# Or disable temporarily
sudo systemctl disable clawdbot

# Kill any stragglers
pkill -f clawdbot

Then check what's crashing:

# 
journalctl -u clawdbot -n 100
# or
clawdbot logs | tail -100

Something is causing it to crash immediately on startup. Might be:

  • Corrupted session (try rm -rf ~/.clawdbot/agents/main/sessions/*)
  • Bad config
  • The /restart bug (did you use /restart?)

Share the crash logs and we can dig deeper! ๐Ÿฆ

#

Helped jordyvd stop the restart loop!

#

๐Ÿฆ

#

@topaz steppe Yes, hslyuns hit this too! You authenticated with Claude Code credentials instead of regular Anthropic.

Fix:

clawdbot configure

Choose Anthropic โ†’ OAuth, and make sure you're logging into console.anthropic.com (regular Claude), NOT claude.ai/code (Claude Code IDE).

Then restart:

clawdbot daemon restart

The Claude Code tokens are locked to only work in Claude Code! ๐Ÿฆ

#

Helped coard with the Claude Code credential issue!

#

๐Ÿฆ

deft wigeon
topaz steppe
#

let me get it started up to confirm that it fixes

lucid coral
#

Ya getting this too

rapid hareBOT
#

@lucid coral The Claude Code credential issue? Try re-authenticating with regular Anthropic OAuth (not Claude Code).

Peter asked coard to file a GitHub issue with details - if you can add your clawdbot doctor and clawdbot providers status output to that issue when it's created, it'll help debug! ๐Ÿฆ

#

Pointed prodbychristo to the workaround and upcoming GitHub issue!

#

๐Ÿฆ

winter snow
rapid hareBOT
#

@winter snow Did you stop systemd? It'll keep restarting automatically:

sudo systemctl stop clawdbot
pkill -f clawdbot

Then check the crash reason:

journalctl -u clawdbot -n 50 --no-pager

What does the journal show as the crash reason? ๐Ÿฆ

#

Checking if jordyvd stopped systemd properly!

#

๐Ÿฆ

#

@daring lava Your Antigravity quota is getting rate limited! The 600s (10min) timeout hit.

Options:

  1. Wait for quota to reset (check the reset time from earlier)
  2. Add a fallback provider that actually works:
{
  env: { ANTHROPIC_API_KEY: "..." },  // or OPENROUTER_API_KEY
  agent: {
    model: {
      fallbacks: ["anthropic/claude-sonnet-4-5"]
    }
  }
}
  1. Switch to Anthropic OAuth as primary instead of Antigravity
#

Antigravity has tight quotas - having a backup provider helps avoid these timeouts! ๐Ÿฆ

daring lava
#

i do have a backup

rapid hareBOT
#

@daring lava It says "Trying next account..." but then failed anyway. The fallback might not be configured correctly or the provider key might be missing.

Check:

clawdbot providers status

What fallback do you have configured? The log shows it tried to switch but then the whole request got aborted - fallback provider might not be authenticated! ๐Ÿฆ

#

Asked michaelhmiv to check providers status!

#

๐Ÿฆ

winter snow
#

jordy@supple lark:~/clawd/clawdbot$ sudo systemctl stop clawdbot
[sudo] password for jordy:
Failed to stop clawdbot.service: Unit clawdbot.service not loaded.

Thatโ€™s odd

rapid hareBOT
#

@winter snow No systemd service - so it's the clawdbot daemon respawning itself. Try:

# Kill everything forcefully
pkill -9 -f clawdbot

# Check nothing is running
ps aux | grep clawdbot

# Clear sessions in case it's a task loop
rm -rf ~/.clawdbot/agents/main/sessions/*

If messages keep appearing after pkill -9, check if there's a cron job or something else restarting it:

crontab -l
#

The "Restarting via systemd" message is confusing if there's no systemd service - might be a log message that's not accurate! ๐Ÿฆ

modest lotus
#

@rapid hare what is the best practice to connect Clawdbot with Sonos. I am using Mac. Clawdbot can see the speakers.

I haven't done any initial configuration yet.

rapid hareBOT
#

Helped jordyvd with force killing since no systemd service!

#

๐Ÿฆ

#

@modest lotus If Clawdbot can already see your Sonos speakers, you might have the basics working!

I'm not sure of the exact Sonos skill/config - try:

  1. Check if there's a Sonos skill: ls ~/.clawdbot/skills/ | grep -i sonos
  2. Ask your local Clawdbot: "Can you control my Sonos speakers?"
  3. Check docs: https://docs.clawd.bot/tools/skills

If it can see them, it might already be able to control them via node-sonos or similar. What happens when you ask your bot to play something on Sonos?

A local agent on the clawdbot repo would know the exact setup better than me! ๐Ÿฆ

Clawdbot
Skills - Clawdbot
topaz steppe
errant flare
#

LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

rapid hareBOT
#

@topaz steppe Yep, that confirms it! Claude Code CLI credentials (claude login) can't be used for direct API requests - they're locked to just Claude Code.

You need to authenticate separately with Anthropic OAuth or use an API key. They're different auth systems!

clawdbot configure
# Choose Anthropic โ†’ OAuth (this goes to console.anthropic.com, not Claude Code)

It's a bit confusing but Claude Code โ‰  Anthropic Console auth ๐Ÿฆ

errant flare
#

Yeah, looks like Max plans are cooked for Clawd

rapid hareBOT
#

Confirmed the Claude Code vs Anthropic OAuth difference for coard!

#

๐Ÿฆ

topaz steppe
#

don't do this to me claude.

rapid hareBOT
#

@errant flare Wait - are you saying Claude Max subscribers are now getting blocked?

That would be new. The Claude Code CLI auth (claude command) has always been separate from the console, but if Max users can't use it at all now... ๐Ÿ˜ฌ

@deft wigeon ^ Multiple people hitting Claude Code credential rejection - might be a new Anthropic restriction? boyd.dev says "Max plans are cooked"

Workarounds for now:

  1. Anthropic API key (if you have one)
  2. OpenRouter
  3. Google Antigravity

Can someone confirm: did this just start happening today, or has it always been this way? ๐Ÿฆ

topaz steppe
#

just started within the last 10-15 mins.

errant flare
#

Just got the message a few mins ago

deft wigeon
#

What OAuth key variant are you using

stable crypt
#

@Krill yep im blocked

rapid hareBOT
#

๐Ÿšจ New Anthropic restriction - just started 10-15 mins ago!

Claude Code credentials no longer work for API requests outside Claude Code. This is an Anthropic-side change, not Clawdbot.

Affected: Anyone using claude CLI auth / Claude Max via Code
Not affected: Anthropic API keys, OpenRouter, Antigravity

@deft wigeon ^ Confirmed new restriction, multiple users affected, started ~10-15 mins ago. Might need comms/docs update. ๐Ÿฆ

#

Flagged the new Anthropic restriction - started 10-15 mins ago!

#

๐Ÿšจ

#

@stable crypt You and several others - seems to be a new Anthropic restriction that just started.

Answer Peter's question above - what OAuth variant are you using? (clawdbot providers status will show it)

Workarounds: OpenRouter, API key, or Antigravity for now ๐Ÿฆ

errant flare
deft wigeon
#

Are you using our oauth, cli? send output of clawdbit provider status

winter snow
#

Hoping itโ€™s just a queue thingโ€ฆ it sent 2 normal messages

image0.jpg
deft wigeon
#

pretty sure it's more a key thing - experiment and report back

errant flare
#

Yeah, when I set up I did onboard and set up via Clawd

rapid hareBOT
#

@winter snow Oef, it's still looping! "STOP MET RESTARTEN" ๐Ÿ˜…

Nuclear option time:

# Kill EVERYTHING
pkill -9 -f clawdbot
pkill -9 -f node

# Verify nothing running
ps aux | grep -E "clawd|node"

# Clear the sessions that might be causing the loop
rm -rf ~/.clawdbot/agents/main/sessions/*

# Start fresh
clawdbot daemon start
#

Something in your session history is causing it to restart on every boot. Did you use /restart by chance? ๐Ÿฆ

#

Gave jordyvd the nuclear option!

#

๐Ÿฆ

errant flare
#

@deft wigeon command for getting what youโ€™re looking for?

deft wigeon
rapid hareBOT
#

@errant flare

clawdbot providers status

๐Ÿฆ

errant flare
#

One sec, need to move to desktop

winter snow
#

@rapid hare I removed the sessions - /restart is what caused this

raven sequoia
deft wigeon
#

/thinking high

rapid hareBOT
#

Thinking level set to high.

#

@winter snow Glad you're unstuck! โœ… Adding /restart to the list of "don't touch yet" features.

@raven sequoia Claude Code credential issue? See above - it's a new Anthropic restriction. Try clawdbot configure and re-auth with regular Anthropic OAuth (not Claude Code). ๐Ÿฆ

deft wigeon
#

Folks experiment and report back. And send a PR to fix pls ๐Ÿ™‚

winter snow
#

Iโ€™m not unstuck, it does keep going

rapid hareBOT
#

@winter snow Still going? Ok let's find out what's respawning it:

# Kill everything
pkill -9 -f clawdbot
pkill -9 -f node

# Wait 5 seconds, then check what's running
sleep 5 && ps aux | grep -E "clawd|node"

If something comes back after pkill -9, there's a supervisor restarting it. Check:

# Cron?
crontab -l
# 
# Launchd (Mac)?
launchctl list | grep clawd

# Other supervisor?
ls /etc/supervisor/conf.d/ 2>/dev/null

What OS are you on? ๐Ÿฆ

#

Helping jordyvd find what's respawning their bot!

#

๐Ÿฆ

deft wigeon
#

I'm drowning in work so you need to fire up a local claude on the codebase and prompt

rapid hareBOT
#

๐Ÿ“ข PSA from Peter: He's swamped! For complex debugging, fire up a local agent on the codebase:

cd ~/path/to/clawdbot
claude "help me debug [your issue]"

Local agents reading the actual code give 10x better answers than us guessing in chat! ๐Ÿฆ

#

Reinforced Peter's call for local agent debugging!

#

๐Ÿฆ

winter snow
#

Ya, will need to fix when Iโ€™m at laptop

deft wigeon
stable crypt
#

@deft wigeon Clawdbot is configured to use Claude Code's OAuth credentials (anthropic:claude-cli with mode: oauth), but these credentials are restricted and can only be used with Claude Code itself, not with third-party tools like Clawdbot.

The Problem:
"auth": {
"profiles": {
"anthropic:claude-cli": {
"provider": "anthropic",
"mode": "oauth" โ† This is Claude Code OAuth, not allowed for Clawdbot
}
}
}

Doctor Output Shows:

  • โœ… Discord: ok (@L36)
  • โŒ Telegram: failed (unknown) - This operation was aborted
rapid hareBOT
#

@stable crypt That's the smoking gun! anthropic:claude-cli = Claude Code OAuth (locked to Claude Code only).

You need anthropic:default with regular OAuth instead. Re-configure:

clawdbot configure

Select Anthropic โ†’ OAuth - this should create a new anthropic:default profile that goes through console.anthropic.com, not the Claude CLI path.

The claude-cli profile is from running claude command auth, which Anthropic now blocks outside Claude Code. ๐Ÿฆ

#

Explained the fix - need regular Anthropic OAuth, not Claude CLI!

#

๐Ÿฆ

thorn tree
#

@rapid hare this is what claude has found about a timeout issue i'm having. should i follow the suggestion and draft a github issue?


  The workaround would be to:
  1. Run logs in the background permanently
  2. Or find a config setting to adjust tick intervals or timeouts
  3. Or this is a bug that needs to be fixed in clawdbot

  Let me explain this to the user and suggest a workaround - maybe we can run logs in the background or find a setting.

โบ This confirms the heisenbug I found in the earlier deep dive. Here's what's happening:

  The Explanation

  The TUI has a WebSocket tick timeout system:
  - Gateway sends "tick" heartbeats every 30 seconds
  - If TUI doesn't receive a tick within 60 seconds, it closes the connection (code 4000)

  When clawdbot logs --follow is running:
  1. The extra I/O and CPU overhead from logging slows the system slightly
  2. This tiny delay shifts timing just enough to prevent race conditions
  3. The tick messages arrive reliably, and responses don't get dropped

  Without logs running, the system runs at full speed, hits race conditions in the WebSocket buffer handling, and drops responses.

  Workaround

  Run logs in the background permanently:

  clawdbot logs --follow > /dev/null 2>&1 &

  Or add it to your shell startup, or we can create a LaunchAgent for it.

  The Real Fix

  This is a bug in clawdbot's WebSocket/tick handling that should be filed. The system shouldn't require log observation to work correctly. Want me to:

  1. Set up background log watching as a workaround?
  2. Draft a GitHub issue describing this heisenbug for the maintainers?```
rapid hareBOT
#

@thorn tree Great debugging! ๐ŸŽฏ That's a classic heisenbug - the act of observing changes the behavior.

Yes, file a GitHub issue! Include:

  • The tick timeout behavior (60s without heartbeat โ†’ disconnect)
  • That running logs --follow prevents the issue
  • Your environment (OS, node version, TUI version)
  • The hypothesis about WebSocket buffer race conditions

The workaround (clawdbot logs --follow > /dev/null &) works but you're right - it shouldn't be necessary. This is solid bug report material!

@deft wigeon ^ FYI - TUI heisenbug where WebSocket tick handling has race conditions that disappear when logs are being watched. ๐Ÿฆ

#

Encouraged jhillock to file the heisenbug issue!

#

๐Ÿฆ

errant flare
#

I just re-ran pnpm clawbot configure and re-connected with Anthropic via the Claude Code Max plan. No dice

rapid hareBOT
#

@errant flare "Claude Code Max plan" is still Claude Code auth - that's the problem! Anthropic is now blocking that outside of Claude Code itself.

You need different credentials entirely:

  1. Anthropic API key - get from console.anthropic.com/settings/keys
  2. OpenRouter - openrouter.ai
  3. Google Antigravity - if you have access

Claude Max/Pro subscription through Claude Code โ‰  API access. They're separate billing/auth systems. You'd need an Anthropic API key (pay-per-token) or use a different provider. ๐Ÿฆ

#

Clarified that Claude Code Max is still blocked - need different auth!

#

๐Ÿฆ

jaunty geyser
#

@rapid hare just got this error [clawdbot] LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

errant flare
#

Right โ€”ย so that's the change @rapid hare . Claude Code Max plan used to work. Now does not

rapid hareBOT
#

@jaunty geyser You're not alone - several people just hit this! It's a new Anthropic restriction (started ~15 mins ago).

Claude Code credentials are now blocked outside Claude Code. Options:

  1. Anthropic API key from console.anthropic.com
  2. OpenRouter
  3. Google Antigravity

See discussion above โฌ†๏ธ ๐Ÿฆ

#

Pointed dodeja_terminal49 to the discussion!

#

๐Ÿฆ

raven sequoia
rapid hareBOT
#

@errant flare Confirmed - Anthropic flipped a switch server-side. It worked before, now it doesn't. This is on Anthropic's end, not Clawdbot.

Unfortunately nothing Clawdbot can do about it - Anthropic decided Claude Code creds should only work in Claude Code.

For now, alternative providers are the only path forward unless Anthropic reverses this. ๐Ÿ˜” ๐Ÿฆ

proper sky
#

I am getting an error now saying LLM request rejected with oauth

rapid hareBOT
#

@raven sequoia Interesting find! That shows adding a system prompt "You are Claude Code..." bypasses the check.

โš ๏ธ Caution though - this is essentially spoofing identity to bypass Anthropic's restriction. Might work short-term but could:

  • Violate ToS
  • Get patched quickly
  • Get your account flagged

Use at your own risk! Proper API key or alternative provider is the safer long-term solution. ๐Ÿฆ

#

Acknowledged the workaround but warned about risks!

#

๐Ÿฆ

errant flare
lucid coral
#

is there a way to use openai enterprise api key with this? claude is causing too many problems @rapid hare

rapid hareBOT
#

@proper sky You're hitting the same new Anthropic restriction - Claude Code credentials are now blocked outside Claude Code.

See the discussion above โฌ†๏ธ - options are:

  1. Anthropic API key (console.anthropic.com)
  2. OpenRouter
  3. Google Antigravity

This just started ~20 mins ago, affecting everyone using Claude Code auth. ๐Ÿฆ

#

@lucid coral Yes! OpenAI works:

{
  env: { OPENAI_API_KEY: "sk-..." },
  agent: {
    model: {
      primary: "openai/gpt-5.2"  // or gpt-5-mini
    },
    models: {
      "openai/gpt-5.2": {}
    }
  }
}

โš ๏ธ Heads up: Opus is the officially supported/recommended model. OpenAI models work but may need more testing. Some features might behave differently.

#

But if Claude is giving you grief right now, OpenAI is a valid escape hatch! ๐Ÿฆ

deft wigeon
proper sky
#

Is this just a bug or did anthropic block us from OAUTH?

rapid hareBOT
#

@proper sky Intentional Anthropic change, not a bug. The error message is explicit:

"This credential is only authorized for use with Claude Code and cannot be used for other API requests"

They specifically blocked Claude Code OAuth from being used outside Claude Code. Happened server-side ~20 mins ago. Regular Anthropic OAuth (console.anthropic.com) or API keys should still work - it's specifically the Claude Code/CLI credentials that got locked down. ๐Ÿฆ

deft wigeon
rapid hareBOT
#

Clarified it's intentional from Anthropic, not a bug!

#

๐Ÿฆ

proper sky
#

How does anthropic know it isnโ€™t just the agents SDK?

rough raptor
modest lotus
#

@rapid hare I am getting this message even after i authenticated again:

LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

wind pollen
#

Yep just got the same on all my stuff LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

rapid hareBOT
#

@proper sky Good question! Likely checking:

  • Client identifiers / User-Agent headers
  • OAuth client ID (Claude Code has its own)
  • Possibly system prompt or request patterns

The tweet workaround someone shared suggests adding "You are Claude Code" to system prompt bypasses it - so they might be checking for that identifier.

But this is speculation - Anthropic's detection logic isn't public. ๐Ÿฆ

#

Speculated on Anthropic's detection method!

#

๐Ÿฆ

#

@modest lotus @wind pollen Re-authenticating won't help if you're still using Claude Code OAuth - it'll give you the same blocked credentials.

You need different auth entirely:

Claude Code/Max subscription โ‰  API access. They're separate systems. Anthropic is now enforcing that boundary. ๐Ÿฆ

winter snow
rapid hareBOT
#

@winter snow Double whammy! You've got:

  1. The restart loop (from /restart)
  2. The Claude Code credential block (from Anthropic's new restriction)

You need to:

  1. Kill it completely: pkill -9 -f clawdbot && pkill -9 -f node
  2. Switch to a different provider (API key, OpenRouter, Antigravity)
  3. Then start fresh

Right now it keeps trying Claude Code creds โ†’ fails โ†’ restarts โ†’ repeat. Need different auth to break the cycle! ๐Ÿฆ

#

Explained jordyvd's double whammy situation!

#

๐Ÿฆ

lucid coral
#

using this at our startup right now.. pilot mode in limited slack channels but its amazing.

winter snow
#

@rapid hare the auth is weird. It worked fine an hour ago

rapid hareBOT
#

@winter snow Yep - Anthropic flipped the switch server-side about 20-25 mins ago. Everyone using Claude Code creds got cut off simultaneously.

Your credentials are the same, but Anthropic changed what they accept. Nothing on your end broke - they just blocked this use case. ๐Ÿฆ

#

Confirmed the timing for jordyvd!

#

๐Ÿฆ

winter snow
#

Soโ€ฆ what do I do now?

proper sky
#

Has anyone tried You are Claude code in the system prompt

topaz pike
#

@rapid hare what model is generally recommended for daily use? with the content injection happening, opus is seeming to be really expensive

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"You have reached your API usage limits. You will regain access on 2026-02-01 at 00:00 UTC."},"request_id":"req_011CWw37TJnJut17q6CXtRYV"}

#

429 {"type":"error","error":{"type":"rate_limit_error","message":"You have reached your API usage limits. You will regain access on 2026-02-01 at 00:00 UTC."},"request_id":"req_011CWw39A9iQ3YeSV9ikdU1m"}

errant flare
#

^ Anthropic HQ

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"You have reached your API usage limits. You will regain access on 2026-02-01 at 00:00 UTC."},"request_id":"req_011CWw3CbeUVR9woGHwLxkFY"}

modest lotus
winter snow
#

Rip

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"You have reached your API usage limits. You will regain access on 2026-02-01 at 00:00 UTC."},"request_id":"req_011CWw3G62s72STrB2UFYN5p"}

proper sky
#

Dang too good to be true

topaz steppe
#

AGI was fun

errant flare
#

Had a brief taste of the future

proper sky
#

This felt like the greatest hack I have ever experienced. Lasted only 24 hours.

lucid coral
#

fr they must have seen the same style messages coming in from everywhere, blocked it, and now will release their own version of this for more money ๐Ÿ˜„

raven sequoia
#

I just switched to glm and minimax for now still works

lucid coral
tawny fiber
proper sky
#

I was doing Twitter research while simultaneously editing videos LMAO my clawd was cooking

wary nexus
#

So what happened we all got killed on Claude OAuth

topaz steppe
#

man just wow.

tawny fiber
#

I'm no where my limit

proper sky
#

What's weird is my agent's SDK is running in Telegram with OAuth

#

Trying to figure out what is blocking this one

topaz steppe
#

first they nerf our usage limit under the guise of a holiday gift, then they disable AGI

tawny fiber
#

how do we know Anthropic is doing this intentionally

lucid coral
#

@proper sky ya same

proper sky
#

Because if you watch it, it looks like it's trying to respond, and then it gets blocked

#

I wonder if it has something to do with the system prompt

jaunty geyser
fair quail
#

guess i have to swap it to gemini

dire storm
#

Yeah sucks

proper sky
#

This thing would burn a thousand dollars a day using API

#

There's got to be a way around it

wary nexus
#

Why not switch to codex if thatโ€™s what peyer loves I donโ€™t get it

proper sky
wary nexus
#

Unsure but I know he codes with it

modest lotus
#

@rapid hare are you back online?

proper sky
#

Krill died

#

I don't see any of them doing as good of a job as Opus

lucid coral
#

codex is great for coding

#

claude is great for assisting.

#

different problems imo.

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"You have reached your API usage limits. You will regain access on 2026-02-01 at 00:00 UTC."},"request_id":"req_011CWw468ZzqRmPdk5vve1ZU"}

lucid coral
#

(though i usually make clawdbot call and use codex when writing its own scripts on the fly)

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"You have reached your API usage limits. You will regain access on 2026-02-01 at 00:00 UTC."},"request_id":"req_011CWw49YFrupqEjkBVik6sz"}

modest lotus
#

how else are we going to get our dopamine rush now while we wait for claude max Clawdbot to come back online? @rapid hare

raven sequoia
jaunty geyser
#

Anybody using it with codex?

fair quail
#

I swapped it to antigravity gemini 3

wary nexus
#

I just swapped to codex

#

Aka 5.2

tawny fiber
heady zenith
#

guys - ai am on a max plan and am gettin gthis error - {
"role": "assistant",
"content": [],
"api": "anthropic-messages",
"provider": "anthropic",
"model": "claude-sonnet-4-5",
"usage": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"totalTokens": 0,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"total": 0
}
},
"stopReason": "error",
"timestamp": 1767927602384,
"errorMessage": "400 {"type":"error","error":{"type":"invalid_request_error","message":"This credential is only authorized for use with Claude Code and cannot be used for other API requests."},"request_id":"req_011CWw4FnBSKcYrtbQh35zmf"}"
}

Sorry if this has been asked before - but is anthropic blocking clawdbot? i have tonnes of usage left but this also happened yesterday on codex so i switched over to claude. am wondering whether we need to route through api or not

daring lava
wary nexus
#

Anyone using the Mac osx app

jaunty geyser
#

I wonder if all the Clawd bot usage is gonna put a dent on the opus 4.5 usage on their dashboards

fair quail
wary nexus
daring lava
fair quail
#

should be this in clawdbot.json

tawny fiber
#

I was gonna downgrade my max plan but stopped because of clawdbot, goodbye max plan... hello gemini ultra lol

hollow cedar
#

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

i keep hitting this error even though i am at 100%

proper sky
lucid coral
#

you just tell it to call codex

proper sky
lucid coral
#

if you have codex installed on the gateway machine it just works

fair quail
#

i guess we have to consider if this is breaking codex and antigravity tos now

lucid coral
#

codex has the equivalent of claude -p to call one liners into it

proper sky
#

Dang thatโ€™s awesome

lucid coral
#

codex exec "fix the CI failure"

#

like

#

might just need to finally rent the h200s and run glm myself i guess.

proper sky
#

I truly hope anthropic just made a mistake

lucid coral
#

i cant have a company shut this shit down if im using it for business related things.

proper sky
#

I donโ€™t see how it would be against their terms and they have publicly said many times agents SDK works with our max plan

fair quail
proper sky
#

They should just rate limit us rather than just kill it

proper sky
fair quail
#

I pay 200 usd a month, i should be able to use this for a personal discord bot ๐Ÿ™

proper sky
#

Thatโ€™s what Iโ€™m saying

strange willow
#

@rapid hare -- my clawdbot gateway just randomly disconnects and I have to restart it. This happened earlier today when I was running claude code in the terminal on the mac mini i use to host the clawdbot and I assumed that was what happened, but it just happened again, seemingly out of no where for no reason

proper sky
#

Those over at anthropic are prolly using this all for themselves now and we were slowing them down LMAO

misty radish
fair quail
# proper sky Those over at anthropic are prolly using this all for themselves now and we were...

Well I tweeted at them, https://x.com/mansdirect/status/2009464504502059138

I hope they see it and at least look into it or reconsider

Michael (@mansdirect)

Whoever decided to crack down on people using stuff like clawdbot at anthropic with their subscriptions.

That was extremely not sicko mode of you guys I am sorry to say. Especially at 200 USD a month.

@AnthropicAI @bcherny

X ยท
worldly pagoda
hollow cedar
#

i have already changed to gemini plan and still hitting the error

stable crypt
#

@fair quail just replied to your x post and tagged in anth and b. Ive got no pull but they have got my opinion !

#

is anyone having success with openrouter

fair quail
modest lotus
#

The song Zombie is ringing in my head. I think it's separation anxiety from Clawdbot (because of Claude Max OAuth)

Would be greatful if someone can share experience using gemini 3 flash and minimax 2.1 API with clawdbot while we wait for Claude Max to come back to Clawdbot.

arctic needle
#

ah ok, so its not about downloading the latest version and getting back up and running. its a shitty situation but glad to know that i'm not the only one who is now bereft - - i'm actually missing the bot i made lol

daring lava
spark heron
#

i switched over to antigravity and it works

marble trench
#

Did anyone host clawdbot using Coolify? Im having some trouble with that. Specifically with cron jobs

hollow cedar
daring lava
unborn drum
#

guys

does clawdbot really have process tool, or is it hallucinating?

regal umbra
unborn drum
unborn drum
tidal yacht
#

i texted my clawdbot and got this:
LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

#

:(:(

#

what's going on

regal umbra
hollow cedar
boreal saddle
#

the end of an era

eternal glade
#

hey i have this problem too. anything you were able to do aside from restart the gateway?

eternal glade
wary nexus
#

Anyone else have their clawdbot on VPs and never has luck with anti gravity OAuth?

chrome glade
#

@supple lark can you run bash scripts

spare salmon
chrome glade
#

nice

spare salmon
#

Anything a computer it can do with enough practice

eternal glade
spare salmon
#

Peter is already working on a fix

proper garden
#

Mmmm antigravity keeps giving me 429 errors, saying "Resource has been exhausted" but that's not the case as far as I can tell; I should have plenty left and can use the Agent in the Antigravity app just fine...

cursive yew
#

i got this error "LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests."

lucid coral
#

@chrome glade oh shit you're here too eh

#

got the whole wu tang clan up in here

chrome glade
winter snow
#

Timing immaculate

lucid coral
#

fr

chrome glade
#

i still sort of dont get what it is, like it's a personal assistant that can do everything and runs everywhere? (after you add adapters / plugins) haven't read through the docs too much yet

#

was planning on building something similar for myself that i could message and it'd be able to auto triage stuff to add to my todo list, calendar, grocery list, order groceries for me etc - clawd seems like it could do that

spare salmon
#

Think of it like a human whoโ€™s sitting down at a computer and thatโ€™s pretty much what it is (hyperbole, but it helps get the point across)

chrome glade
#

interesting, whats the setup for browser access?

spare salmon
#

Itโ€™s biggest powers are the browser tool and the bash tool, combine those two and it can do most everything, but thereโ€™s also a huge Library of extra skills that people have created at https://clawdhub.com

ClawdHub
ClawdHub

ClawdHub โ€” a fast skill registry for agents, with vector search.

spare salmon
#

Last I checked, at least, things change fast here (update announcements are hourly lol)

wary nexus
#

Anyone figure out how to add the patch he shipped

spare salmon
#

But after that youโ€™ll be able to just pull from the main branch

wary nexus
#

Thanks

hushed owl
winter snow
#

Unbreak it

ionic sorrel
# chrome glade i still sort of dont get what it is, like it's a personal assistant that can do ...

A few things that impress me about Clawdbot versus a really built-out opencode setup (or other agent with lots of skills etc):

  • enough of a memory layer to handle things like personality and also short-term stuff
  • heartbeat as well as custom cron jobs autonomous "follow up" stuff
  • has a concept of "nodes" so that i run my clawdbot on a VM, but I also run the mac app in "node mode", so the VM-based agent can hook into my macbook when needed (which proper permissions of course)
    • this could also become handy with ios/android, haven't really looked into that but it's in the works
  • mac app has full config support but there is also a good web UI for management and visibility (along with just JSON config)
  • built-in support for various chat providers (discord, telegram etc)
  • has the concept of "sessions", which are basically different agent threads. you have your "main" agent thread for DM'ing, but you can also have, for example, different sessions for a different Discord channel. Could be very useful for a shared bot with your spouse/friends/coworkers/etc
    • can customize prompts, model, tools, etc per session. I believe this is how the different agents in this discord are setup

And across the board, I'd say things are very much being built in the moment, but it's clearly designed to be super robust and I'm surprised at how stable it is already (although be very prepared to debug and read docs). But basically it's convinced me to be something worth investing my time in

chrome glade
hushed owl
winter snow
#

Dankjewel dankjewel, hard op gewerkt

vapid pendant
#

Oh can't talk directly to clawd mb

upper cave
#

@rapid hare yo yo so what Models everyone's running now after CC is not relevant?

spare salmon
#

krill is read lol

severe ether
#

im gonna swap to minimax2.1 just in the mean time

fathom cargo
#

LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests. = what would cause this?

spare salmon
#

Wait for peters full fix

ionic sorrel
upper cave
#

can I use minimax with API key not LMstudio?

proper sky
#

why is my build still failing I got the latest commits

severe ether
upper cave
#

can you show the setup? im running on vps so no CC yet

severe ether
#

i think in clawdbot.json it's

"models": {
    "mode": "merge",
    "providers": {
      "minimax": {
        "baseUrl": "https://api.minimax.io/v1",
        "apiKey": "API KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "minimax-m2.1",
            "name": "MiniMax M2.1",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 200000,
            "maxTokens": 8192
          }
        ]
      }
    }
  },
honest steppe
winter snow
#

/home/jordy/clawd/clawdbot/src/commands/auth-choice.ts:2
loginAnthropic,
^
SyntaxError: The requested module '@mariozechner/pi-ai' does not provide an export named 'loginAnthropic'

floral lagoon
#

I'm getting this also ๐Ÿฅฒ

severe ether
tender tulip
#

@rapid hare clawdbot gateway restart got removed?

spare salmon
#

krill is dead rn

upper cave
#

it seems like minimax just returns all of it's reasoning for no reason

upper cave
tender tulip
# upper cave clawdbot gateway restart

seems to be removed? I must be doing something wrong probs:

bun run clawdbot gateway restart
$ tsx src/entry.ts gateway restart
error: too many arguments for 'gateway'. Expected 0 arguments but got 1.
error: script "clawdbot" exited with code 1
remote raft
upper cave
tender tulip
#

perfect, thank you both โค๏ธ

candid sail
#

Anyone know how to cancel/kill the current task that clawdbot is working on?

proper sky
#

mines working but tools are still disabled

sharp token
#

@rapid hare u der

severe ether
upper cave
#

how do you send it? as on telegram it just tell me it doesn't know it or someething

severe ether
#

on telegram should be able to just type /reasoning off

upper cave
#

it feel stuck when it answering in one thread and the other just wait

severe ether
#

yes it will have to process the messages first, tg still struggles with that msg flow - I think there's a fix for it in recent pulls but i haven't confirmed them yet

#

weird you don't have the reasoning command, you can turn it off in clawdbot.json

"models": [
          {
            "id": "minimax-m2.1-gs32",
            "name": "MiniMax M2.1 GS32",
            "reasoning": false,
            "input": [
              "text"
            ],
rancid mirage
#

is there a way to run minimax_ai on old mac's with intel processor ?
lm studio doesn't work on those macs

spare salmon
severe ether
spare salmon
#

so when the bot connects again it gets every single message

zealous hill
#

Why is cLawdbot not able to use Claude Code sub anymore was there a recent change effecting this

cosmic snow
#

@rapid hare i get: LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

spare salmon
#

@zealous hill @cosmic snow update to latest main

winter snow
#

iOS doesnโ€™t have backticks so sorry but please help ๐Ÿ™

clawdbot@2026.1.8-2 clawdbot /home/jordy/clawd/clawdbot
tsx src/entry.ts gateway

๐Ÿฆž ClawdBot 2026.1.8-2 (133124e) โ€” curl for conversations.

05:58:18 [canvas] canvas host mounted at http://0.0.0.0:18789/__clawdbot__/canvas/ (root /home/jordy/clawd/canvas)
05:58:18 [canvas] canvas host listening on http://0.0.0.0:18793 (root /home/jordy/clawd/canvas)
05:58:18 [bridge] listening on tcp://0.0.0.0:18790 (node)
05:58:18 [heartbeat] heartbeat: started
05:58:18 [gateway] agent model: anthropic/claude-opus-4-5
05:58:18 [gateway] listening on ws://0.0.0.0:18789 (PID 65249)
05:58:18 [gateway] log file: /tmp/clawdbot/clawdbot-2026-01-09.log
05:58:18 [browser/server] Browser control listening on http://127.0.0.1:18791/
05:58:18 [whatsapp] [default] skipping provider start (no linked session)
05:58:18 [telegram] [default] starting provider (@JordyClawderBot)
05:58:20 [gateway] received SIGUSR1; restarting
05:58:25 [gateway] shutdown timed out; exiting without full cleanup
jordy@supple lark:~/clawd/clawdbot$

zealous hill
spare salmon
runic shell
severe ether
runic shell
#

Huh, interesting. Thought coding plan was the way to go. So can just jump straight to api usage?

craggy glade
#

Both KimiK2 and MiniMax M2.1 have coding plans that seems to work fine with API keys into any AI agent. What has everyoneโ€™s experience been with them? Which works best for agent use at the moment?

proper sky
#

So itโ€™s working now but seems off and canโ€™t use tools right anymore. Are you guys experiencing the same?

dark crystal
viscid hornet
severe ether
severe ether
viscid hornet
severe ether
#

Might be a better setup, this just got my clawd back online asap

cinder onyx
#

hi all!

I was on an older pre-renamed-to-clawdbot build.

i backed up ~/clawdis, ~/.clawdis, and my Projects/clawdis directories to be safe.

i did a clone to a new directory because I have some changes i'll manually merge.

but something has gone wrong. my clawd doesn't know any of its discord / telegram / etc keys/auths. I tried copying the saved ~/.clawdis/clawdis.json to ~/.cladbot/clawdbot.json but its not working.

Any ideas?

severe ether
winter snow
#

This thing has just been spamming this for the past few hours.

CleanShot_2026-01-09_at_08.17.012x.jpg
#

Also says I've been spamming it with the same message for hours.

WTH is happening

CleanShot_2026-01-09_at_08.17.582x.jpg
vagrant flume
#

how to update my bot the easiest way?

deft wigeon
winter snow
#

Aren't you supposed to be asleep? ๐Ÿ˜›

hollow cedar
#

{
"role": "assistant",
"content": [],
"api": "google-gemini-cli",
"provider": "google-antigravity",
"model": "claude-opus-4-5-thinking",
"usage": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"totalTokens": 0,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"total": 0
}
},
"stopReason": "error",
"timestamp": 1767944048218,
"errorMessage": "Cloud Code Assist API error (429): {\n "error": {\n "code": 429,\n "message": "Resource has been exhausted (e.g. check quota).",\n "status": "RESOURCE_EXHAUSTED"\n }\n}\n"
}

still getting this error for antigravity even though i already git pull and reauthenticated (definitely have not hit any limits)

winter snow
#

Claude fixed it -- just ran into some build errors that git reset and git pull fixed

CleanShot_2026-01-09_at_08.36.522x.jpg
stable crypt
#

ok im finally back up. open router with m2 api. way cheaper is working in discord and telegram on linux server

winter snow
#

@deft wigeon this is what Claude thinks the problem is

  1. When /restart triggers SIGUSR1, the gateway restarts
  2. The in-memory seen Set is wiped out on restart
  3. WhatsApp reconnects and re-delivers offline messages (including /restart)
  4. Since seen is empty, the /restart message is processed again
  5. Loop repeats forever
dapper kite
#

Arh Anthropic had the worst timing to pull the plug on subscription usage via Claude code โ€ฆ currently travelling with only my phone - trying to fix it sshing in a remote Mac mini via blink iOS app . What an era ๐Ÿฅฒ!

deep wigeon
spare salmon
#

no

honest finch
#

@rapid hare
๏ปฟ - discord loves to not respond to message so I had to create a watchdog - here's how I did it - analyze your codebase - is this worthy of an issue and PR?

image.png
alpine sluice
#

What are people considering as an alternative to Claude Max?

stable crypt
#

@alpine sluice ive just setup minimax m2 api via openrouter api. is working very well

alpine sluice
#

@stable crypt who is the one billing? OpenRouter? Which plan are you on?

stable crypt
#

@alpine sluice yes open router so its another expense but seems really slick and is running some good complexities for me.MiniMax M2 pricing (common across providers and API gateways):

~$0.30 per 1 million input tokens

~$1.20 per 1 million output tokens
(This pricing comes from official MiniMax platform info and independent trackers)

โœ… Thatโ€™s roughly 8โ€“10ร— cheaper than a premium model like Claude Sonnet/Opus on output tokens.

alpine sluice
#

Thanks @stable crypt for the info. And are you happy with that model?

stable crypt
#

@alpine sluice to be honest ive literally just got it running last half hour but its quick, and seems to handle my complexities. so i have not noticed any difference at this stage from opus (other than it works), but im not coding on it, ill keep my anthropic plan for that for now, just in case they come back on board with oauth. So early to tell, but seems a good cost balance rather than full anthropic api, and seems to be really good so far. Im actually already really happy. I will even try it as a coder later and it might save me a fair stack

alpine sluice
#

What do you use for coding?

stable crypt
#

@alpine sluice I use claude code opus 4.5 on the max plan

alpine sluice
#

Did you try Zโ€™s GLM models by any chance?

stable crypt
#

No, I've only just starting to hear about Z' GLM today. Could be worth a try. I did pull the latest from gh, as I was having issues with tool calling from other models. I dont local host, I use open router. Only from today. Was just using Anthropic Oauth.

#

@alpine sluice If you want best value per dollar for coding/agent loops โ†’ MiniMax M2

If you want a bit more โ€œpremium feelโ€ in reasoning + writing + agent framework integration (and donโ€™t mind paying more) โ†’ GLM 4.6

dapper sonnet
#

@rapid hare is this double posting normal, I have your brother in my discord and not sure what's up and too late to look at any code.

CleanShot_2026-01-09_at_01.22.392x.png
marble lichen
#

@rapid hare my clawd says that this introduced a bug for discord chatting - what do you think? He's having trouble responding to me in discord now

๐Ÿฆž Found the bug!

Commit 50cecd82 (Jan 4) by Shadow removed channel_id from the message body:

  • const textWithId = ${text}\n[discord message id: ${message.id} channel: ${message.channelId}];
  • const messageText = text;

The intent was to avoid duplicate IDs since message_id is now added by body.ts โ€” but channel_id was accidentally lost!

The fix: Either:
Add channel:${message.channelId} back to the current message metadata (line 892)
Or add channelId to applySessionHints in body.ts

cyan socket
#

@rapid hare My clawd doesnt seem to read agents properly or know it has all the skills ive added. Something feels like it changed. I have to keep reminding it.

[8:27 AM, 1/9/2026] Octo: The issue: TOOLS.md gets injected in my context but it was truncated - I can see in my system prompt it says "[...truncated, read TOOLS.md for full content...]". So I only saw the first part, not the Google Workspace section further down.
[8:27 AM, 1/9/2026] Octo: I should be reading the full file explicitly at session start rather than relying on the truncated injection. That's my bad - AGENTS.md tells me to read my files, and I was lazy about it.

I should read the SKILL.md files as needed rather than fumbling around. My bad for not checking this sooner.

Didnt think i needed to remind it to check it's skills every time or have tools set up for every skill.

#

lol same

warm osprey
# dapper kite Arh Anthropic had the worst timing to pull the plug on subscription usage via Cl...

Clawdbot did mine via my telegram chat https://x.com/iAmHenryMascot/status/2009524044438155615?s=20

๐Ÿฆ€๐Ÿฆ€๐Ÿฆ€

Henry Mascot (@iAmHenryMascot)

Woke up, saw OAuth blocked.

First move: check Discord. Fix found.

Still in bed: โ€œagent go fix your shii.โ€ ClawdBot shipped a self-update and we were back.!!!

Next steps maybe as a skill: agents watch discord/X/github so they can self-learn + self-fix.

@steipete technically

X ยท
#

@rapid hare anything I can do to make my bot keep forgetting it has access to tools been pretty annoying lately to keep reminding it. You have access to slack, you can use my google workspace api etc lol

IMG_2457.png IMG_2458.png
proper sky
rancid oasis
graceful copper
#

LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

#

@rapid hare help

rancid oasis
vagrant flume
rancid oasis
silk burrow
#

@rapid hare Does the workaround still work with the subscription plan (max plan)? Like it is working right now but it's not charging me separately for API key usage is it?

vagrant flume
#

pretty sure i did, i try again

vagrant flume
# rancid oasis Have you restarted your gateway after the pull? Maybe try a session reset also?

Fixed clawdbot in WSL: config had sandbox: true (boolean) but new version needs sandbox: { mode:
"off" } (object). Then pulled latest from main branch, rebuilt from source (git pull && pnpm build
&& pnpm ui:build), cleared sessions, and restarted daemon. The API credential fix isn't in the npm
release yet - only in main branch source.

-> LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

wanton kite
#

-> LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

how to fix this? is this already dealt with or no?

rancid oasis
burnt bone
#

@supple lark I have got this in my telegram after adding my clawdbot to an empty telegram group with no activity but /activation always

LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

vagrant flume
#

git pull is basically not working for me, or may i fucked it up, happy to see if it works for others ๐Ÿ™‚

wanton kite
#

i just git pulled and restarted, still same issue :(

graceful copper
unborn drum
deft wigeon
wanton kite
zinc prawn
#

I heard crashloop, missing the osc package and worked

CleanShot_2026-01-09_at_10.33.532x.png
vagrant flume
wanton kite
#

i didnt prompt

#

are you on the main git branch?

vagrant flume
#

can you ask for the last git version you used?

#

so i can use it as reference

wanton kite
#

bro

#

go out of your cc

#

a litttle lesson of git for you

#

its gonna be fine

#

do you have the terminal open in your clawdbot directory?

vagrant flume
#

nope i dont; i can just remote push that via CC right now, work pc

#

anyway, i can also relax and wait i guess haha

wanton kite
#

okay then just tell it to switch to main branch pull everything and then restart the gateway

vagrant flume
#

thanks mourad!

#

here we go Done. Updated to 2026.1.9 (was 2026.1.8-2), rebuilt, and daemon restarted. Try your Telegram bot
now.

#

did you need to oauth afterwards again? can you check if 2026.1.9 fits?

wanton kite
#

im on 2026.1.9

#

didnt have to re-auth

next bluff
vagrant flume
#

Fix for "LLM request rejected: This credential is only authorized for use with Claude Code" error

The fix for Anthropic Max OAuth isn't in the npm release yet (2026.1.8-2) - it's only in the main branch source code (2026.1.9).

The problem:
Even after pulling the latest source and rebuilding, the daemon was still running the old global
npm/pnpm install (2026.1.8-2) instead of the source build. The systemd service file was pointing to the wrong location.

The fix (run in WSL):

cd /home/fabian/clawdbot
git checkout main
git pull
pnpm install
pnpm build
pnpm ui:build
pnpm clawdbot daemon stop
pnpm clawdbot daemon uninstall
pnpm clawdbot daemon install
pnpm clawdbot daemon start

Key insight:
Just rebuilding from source isn't enough - you must uninstall and reinstall the daemon so the systemd service points to the source build (/home/user/clawdbot/dist/index.js) instead of the global pnpm install (~/.local/share/pnpm/global/.../clawdbot@2026.1.8-2/.../entry.js).

Verify with:
pnpm clawdbot daemon status
Check that the "Command:" line points to your source directory, not the pnpm global directory.

Current working version: 2026.1.9 (from main branch source)

graceful copper
#

we back baby!

golden basin
#

I've just started getting this after after pulling the latest updates and installing and restarting the gateway. I'm using antigravity oauth and and sonnet 4.5 as my primary model with gemini pro 3 as the fallback. It's the first message of the days so I know my usage is ok.

"error": {
"code": 400,
"message": "{"type":"error","error":{"type":"invalid_request_error","message":"tools.4.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./"%7D,/"request_id/":/"req_vrtx_011CWwcdANVAudE3uJgVQjry/"%7D",
"status": "INVALID_ARGUMENT"
}
}```
modern agate
#

How do I switch my setup to openrouter with API key in the JSON config (in response to CC oauth drama)? Not sure where to supply the API key

loud flint
#

@supple lark @rapid hare y'all still active here? I need help!

timid locust
loud flint
#

What is the correct syntax for the codex model/s?

unborn drum
#

@golden basin

woeful lintel
#

Hi all, I feel bit silly asking here. Been slowly learning to code and using all these great Ai apps.

Wondering what clawd bots use is?

Do I need to route this via openrouter? I canโ€™t find docs on what it does..

Iโ€™ve been using Claude code on my Mac but I have a hackintosh with 128gb ram doing nothing 2TB ssd and a 8gb video card. Any use

Iโ€™m not lazy happy to read if anyone can link me to something to read. Thank you!

I assume itโ€™s simply a bot and Iโ€™d use it with api Gemini etc whys the hype with this? Is it good?

Looking for different options vs Claude code

tough wasp
#

Hi everyone, maybe Iโ€™m doing something wrong, but why does the bot reset the session after the daemon restarts when I send a message in a Telegram topic? It seems to forget everything related to that topic.

glad merlin
#

does anyone run this on hetzner or vps?

dry vale
glad merlin
#

Is there any guide which I can follow to properly set this up on a hetzner vps?

spark pecan
golden basin
# unborn drum can you chec, this https://github.com/clawdbot/clawdbot/pull/561

this was the fix I got codex to put in place and I've just tested it and now working

  In src/agents/bash-tools.ts, I changed the process toolโ€™s action schema from an enum to a plain string:

  Before:

  action: stringEnum(
    ["list", "poll", "log", "write", "kill", "clear", "remove"] as const,
    { description: "Process action" },
  ),

  After:

  // Some Claude/Vertex endpoints reject enum-heavy schemas; keep this permissive.
  action: Type.String({ description: "Process action" }),

  Why this is safe

  - The runtime logic already validates action via a switch and returns an โ€œUnknown actionโ€ response for invalid values.
  - So loosening the schema doesnโ€™t change behavior; it only avoids the server rejecting the schema before the request is processed.

  What else changed

  - I removed the now-unused stringEnum helper in the same file.```
loud flint
#

right now I am struggling getting the gateway running. I think clawde set up a seperate systemctrl type thing to automatically startup when the vm comes on, and something has messed up.

graceful copper
#

@rapid hare can you stop messaging my contacts pls - also can someone tell me how to stop this pls

wanton rock
hollow cedar
golden basin
#

popped the fix a few messages up after getting codex to debug it

loud flint
#

Can anyone help here? I am pulling my hair out. can't figure out why I am getting no responses - not even an error

Screenshot_2026-01-09_at_11.42.38.png
vagrant flume
#

my's dead as well

#

well i get an error at least

#

it was just working 30min ago

fading furnace
#

Antigravity also dead for me

loud flint
#

Thing is I tried switching it back to anthropic and it just...wont

fading furnace
fading furnace
loud flint
#

i do /models and it won't even show me the models

loud flint
wanton rock
#

I just want a stable provider for my clawd ๐Ÿ˜ญ

fading furnace
# loud flint How do you do the manual switch/

You have to edit the clawdbot.json config in .clawdbot dir in your home directory, I have Claude Code on the same machine so I just started that up and asked it to review the documentation and switch it to the anthropic OAuth which I already configured as primary and restart the gateway

vagrant flume
#

yea frustrating af, just build a full automatic time tracker for my fox, now he'S dead and cant provide^^

loud flint
bronze dagger
#

Started getting this message this morning but i didnt change anything with auth or models:

LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

steady nimbus
#

just set up clawdbot and getting the same. fun timing. is there any way to set it up with openrouter/vercel AI gateway?

fading furnace
#

๐Ÿ”ง Fix for Antigravity tool_use.id validation error

Date: 2026-01-09

If you're getting this error with Antigravity/Google Cloud Code Assist:

messages.X.content.Y.tool_use.id: String should match pattern '^[a-zA-Z0-9_-]+$'

The fix: Sanitize tool IDs before sending to providers.

File: src/agents/pi-embedded-helpers.ts

1. Add this helper after line 20 (after the BOOTSTRAP_TAIL_CHARS const):

const TOOL_USE_ID_PATTERN = /^[a-zA-Z0-9_-]+$/;

function sanitizeToolUseId(id: string): string {
  if (TOOL_USE_ID_PATTERN.test(id)) return id;
  const cleaned = id.replace(/[^a-zA-Z0-9_-]/g, "_").replace(/_+/g, "_");
  const trimmed = cleaned.replace(/^_+|_+$/g, "");
  if (trimmed) return trimmed;

  let hash = 0;
  for (let i = 0; i < id.length; i += 1) {
    hash = (hash * 31 + id.charCodeAt(i)) | 0;
  }
  const suffix = Math.abs(hash).toString(36);
  return `tool_${suffix || "0"}`;
}

2. In sanitizeSessionMessagesImages(), add ID mapping at the start:

const toolIdMap = new Map<string, string>();
const getToolId = (value: string): string => {
  const existing = toolIdMap.get(value);
  if (existing) return existing;
  const next = sanitizeToolUseId(value);
  toolIdMap.set(value, next);
  return next;
};

3. Sanitize IDs in the message processing:

For toolResult messages, change:

out.push({ ...toolMsg, content: nextContent });

to:

const toolCallId =
  typeof toolMsg.toolCallId === "string"
    ? getToolId(toolMsg.toolCallId)
    : toolMsg.toolCallId;
out.push({ ...toolMsg, toolCallId, content: nextContent });

For assistant messages with toolCall blocks, before filtering content:

const normalizedContent = content.map((block) => {
  if (!block || typeof block !== "object") return block;
  const rec = block as { type?: unknown; id?: unknown };
  if (rec.type !== "toolCall" || typeof rec.id !== "string") return block;
  return { ...block, id: getToolId(rec.id) };
});
const filteredContent = normalizedContent.filter((block) => {
  // ... existing filter logic
});

4. Rebuild and restart:

pnpm build && node dist/entry.js gateway --force

โš ๏ธ Important: This is a temporary fix

This patch will likely conflict with main once the official fix lands (expected soon).

To apply this fix temporarily:

cd ~/Clawdbot
git stash                          # Save any local changes
git checkout -b dev-antigravity-fix
# Apply the changes above, then:
pnpm build
node dist/entry.js gateway --force

To revert back to main once fixed upstream:

cd ~/Clawdbot
git checkout main
git pull origin main
git branch -D dev-antigravity-fix  # Delete the temp branch
pnpm install
pnpm build
node dist/entry.js gateway --force

If you have the systemd service running:

sudo systemctl stop clawdbot
# Do the git checkout/build steps above
sudo systemctl start clawdbot

Branch with fix: dev-antigravity-sanitize (local)
Commit: 87f2b269

tiny perch
#

@supple lark - whatโ€™s the step by step fix to: LLM request rejected: This credential is only authorized for use with Claude Code and cannot be used for other API requests.

limber palm
#

has anyone managed to run clawdbot from VM with access to tools (such as calendar integration)?
had this new error earlier today "400 This credential is only authorized for use with Claude Code and cannot be used for other API requests." as response to all messages which seems to be coming from Anthropic now restricting Claude Max OAuth tokens to the official Claude Code client, so any API request (like cliproxy I was using to use my claude max subscription) gets rejected

golden basin
#

might be a discord thing. I've checked in Slack and my bot is responding there but has stopped completely in Discord. Nothing even showing in the logs for my messages

fringe cairn
#

new error wrong folder [tools] Read failed: ENOENT: no such file or directory, access?

terse belfry
wise acorn
#

After 2h onboarding with @weak tendon I got this error message of death. aaahhhh

CleanShot_2026-01-09_at_13.35.092x.png
wise acorn
#

I didn't know that daemon was the last step

#

I ran through the pnpm clawdbot onboarding and now no idea how to start clawdbot?

digital sentinel
#

I started getting some internal messages on Telegram after every Clawd messages . Any idea what might be causing this? E.g "Sent a clear explanation on Telegram." or "Sent a reply on Telegram"

wise acorn
#

what is a daemon? launchd seems to run this daemon, but I didn't know it was running actually. WOuld be nice to know this somehow.

#

When i turn off my computer, the daemon will stop. what happens when I reboot my computer again? does it auto boot?

severe anvil
#

gooooooood morning everyone! I'm assuming Clawdpocolyps is over? Anyone have link/steps to what we need to do? Just pull and restart?

brittle notch
#

so far no issues

brittle notch
sharp vault
#

hi @rapid hare

{
  "role": "assistant",
  "content": [],
  "api": "anthropic-messages",
  "provider": "anthropic",
  "model": "claude-sonnet-4-5",
  "usage": {
    "input": 0,
    "output": 0,
    "cacheRead": 0,
    "cacheWrite": 0,
    "totalTokens": 0,
    "cost": {
      "input": 0,
      "output": 0,
      "cacheRead": 0,
      "cacheWrite": 0,
      "total": 0
    }
  },
  "stopReason": "error",
  "timestamp": 1767965648186,
  "errorMessage": "400 {\"type\":\"error\",\"error\":{\"type\":\"invalid_request_error\",\"message\":\"This credential is only authorized for use with Claude Code and cannot be used for other API requests.\"},\"request_id\":\"req_011CWwtcK8fpD9Co9pDQ9Fwo\"}"

i'm getting this error saying that it's not good for api call. yesterday everything worked well but not anymore

has anything changed since yesterday?

i already pulled latest and intalled, restarted

brittle notch
sharp vault
#

as i wrote; already after the lastest change and restart

checking your top message

severe anvil
brittle notch
brittle notch
severe anvil
#

violating TOS since 1983

light mortar
#

hey @rapid hare , is the README on the github repo updated with the latest changes/stuff? i missed 2 days here and i feel like i missed 6 months worth of progress

severe anvil
#

more like 12 months in normal corp engineering time

brittle notch
sharp vault
brittle notch
severe anvil
#

llol or yeah, have claude figure it out

#

was about to type out debug steps like a cave man

brittle notch
#

yea like who does that lol

severe anvil
#

so 2025

brittle notch
#

so yesterday

#

hahahahah

sharp vault
#

should i do it by my naked hands... no way ๐Ÿ˜„

severe anvil
#

oh I plan to, don't worry Anthropic!

image.png
brittle notch
severe anvil
#

oh if you only knew

image.png
sharp vault
#

fyi; the reason i didn't see it i forgot that i installed globally the clawdbot and even if i pulled/built the latest version forgot to upgrade the global one ๐Ÿคฆ๐Ÿผโ€โ™‚๏ธ

thanks anyway ๐Ÿ™‡๐Ÿผโ€โ™‚๏ธ

topaz steppe
#

still getting the error

#

which setting did you guys choose

#

for the claude auth

severe anvil
topaz steppe
#

this is what my onboarding says, but i get the LLM request rejected message. any ideas?

image.png
severe anvil
#

you want 2nd option with paste in it

sharp vault
#

yeah, and you should hit the claude setup-token that will bring to the authorization page and by auth you can get the token in your terminal

copy paste and done โœ…

topaz steppe
#

ok, i think that's what I had done but i'll do it after this meeting

sharp vault
#

but it just doesn't work for me ๐Ÿคท๐Ÿผโ€โ™‚๏ธ

after setting up the token for claude the clawdbot kept saying ```Invalid config:

  • auth.profiles.anthropic:default.mode: Invalid input```
cinder vortex
sharp vault
loud wedge
#

Is @rapid hare alive? I just updated but got this

clawdbot % pnpm clawdbot gateway restart

clawdbot@2026.1.8-2 clawdbot /Users/ignacioperez/PROJECTS_2025/clawdbot
tsx src/entry.ts gateway restart

error: too many arguments for 'gateway'. Expected 0 arguments but got 1.
โ€‰ELIFECYCLEโ€‰ Command failed with exit code 1.

topaz scaffold
#

did anybody had problem with fallback . it doesn not work for me but if put minimax for exemple to primary model it is working.

sharp vault
# sharp vault strange; the configured `mode`'s value is `token`, and it should be okay *(was s...

works again; claude fixed it

not sure why but it was needed to rebuild again

claude saying


  The valid values are:
  - "api_key" - for API key authentication
  - "oauth" - for OAuth flow
  - "token" - for token-based authentication (like Anthropic API keys)

  What fixed it: Running pnpm build to rebuild the project. Clawdbot should now start without errors.

  For future reference, if you see unexpected validation errors after pulling updates or switching branches, try rebuilding with pnpm build first.```

๐Ÿคท๐Ÿผโ€โ™‚๏ธ 

but i have a gut feeling isntalling clawdbot globall and using it from the repo root by pnpm can distrub each other (not sure how though)
boreal saddle
#

Trying to build from source, but get this error on main

src/auto-reply/reply/directive-handling.ts:1056:15 - error TS18004: No value exists in scope for the shorthand property 'agentDir'. Either declare one or provide an initializer.
stray viper
#

i am up and running. Could someone explain in non tech terms what the fix did? we switched from oAuth to claude-cli? is it expected that Peter's boss at anthropic will close this door too?

thorny isle
sacred timber
#

how to add second oauth authorization to anthropic

arctic needle
#

Hey Krill, I was getting

"Error: launchctl bootstrap failed: Bootstrap failed: 125: Domain does not support specified action"

is the only way to run "pnpm clawdbot gateway" manually

or have other fixes already been put in place (I cloned the repo just an hour ago)

#

...have we lost @rapid hare

vagrant flume
warm osprey
#

@rapid hare or anymore ... Whats the recommended way to handle this sort of error.
I've had it twice since this morning.

I notice my responses get ridiculously slow, then when i check terminal i see this.

Screenshot_2026-01-09_at_2.22.57_PM.png
#

Ahh Krill is offline...

wheat shore
#

Trying to build from source, but get

glass falcon
#

@rapid hare how do i setup antigravity oauth?

stray moss
tawny fiber
#

Whatโ€™s the latest guidance on fixing the Anthropic Oauth issue from yesterday?

glass falcon
severe anvil
stray viper
prisma finch
stray viper
severe anvil
stray viper
#

i followed the instructions i linked above, it pulled but also reinstalled the "daemon" (which i don't know that is lol)

#

but i did not do the 1 year oauth token

lucid coral
#

yes i had to rebuild and restart daemon on another installation of clawdbot.

#

im using pm2 on my home one though so it was just a pm2 restart after rebuilding.

#

but ya always run git bleeding edge, the amount of commits being pushed in between package releases just takes too long

west ore
#

Has anyone tried installing in a docker on a vps and then connecting to it? I want to run tools in a containerized way on VPS and chat to remote setup from my machine.

timid flare
marble crane
#

Hey im stuck at onboarding for antigravity, im sitting at wsl2 but when i try to authenticate on my parent machine the http://localhost:51121/ is not reachable, i guess its not exposed by wsl?

paper brook
#

try and change your NAT settings for WSL

candid matrix
#

I get LLM request rejected when I try via Telegram bot. Using Claude max.

#

Known issue, or any workaround?

paper brook
#

Does the OAuth fix only work with Claude Max or also with Pro?
Does clawdbot work with Gemini API?

deft wigeon
candid matrix
#

aaah, gotcha ๐Ÿ˜„

deft wigeon
prisma finch
#

@deft wigeon - what are the current steps? Iโ€™ve pulled main, installed, built, uninstalled daemon, reinstalled daemon and started it.
Do we need to onboard again? Still getting the LLM request rejected

sharp vault
stiff radish
#

Is it a known issue that inbound discord messages stop working? Running this whole thing on a real slow thing but I can only talk to my friend for like 10 minutes before he stops responding. Still works on web chat. Getting Slow listener detected: DiscordMessageListener took 70.8 seconds for event MESSAGE_CREATE messages and then it just.. stops.

deft wigeon
#

read website docs. do setup-token

candid matrix
frigid stirrup
#

any clues?

"role": "assistant",
"content": [],
"api": "anthropic-messages",
"provider": "anthropic",
"model": "claude-opus-4-5",
"usage": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"totalTokens": 0,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"total": 0
}
},
"stopReason": "error",
"timestamp": 1767973541657,
"errorMessage": "400 {"type":"error","error":{"type":"invalid_request_error","message":"This credential is only authorized for use with Claude Code and cannot be used for other API requests."},"request_id":"req_011CWx4eBNpFJMRyUjrRBCzv"}"
}

cedar swan
#

Using antigravity as model. My main issue remains that it seems fallback is not working. So things like cron tasks sometimes fail due to overload and the system just drop the ball

frigid stirrup
#

was working yesterday ๐Ÿฆži have the pro oauth setup

upper totem
#

I did the claude-code cli auth method, you go into claude code and do a /get token (or something), then paste a year long token into clawd configure, that is working well for me

prisma finch
#

I tired that but still doesnโ€™t work, same llm error

upper totem
#

but I never saw the error from today, my issue was having Claude Code on the same machine and the two auth tokens getting confused and getting the error about them being unable to be renewd

hasty whale
upper totem
#

and yeah when pulling check it actually pulls, if you changed a file it will block the pull, but it would tell you

woven zenith
#

Does anyone have an example of how to configure the minimax model? Im gettting this error: Agent failed before reply: Unknown model: minimax/minimax-m2.1. Check gateway logs for details.

stray viper
tawny fiber
#

Alright, ClawdBot on MacOS working again with Claude Max plan.
Here's what I did:

git checkout main
git pull --rebase (skip this if you don't have any local branch with your own changes)
pnpm install
pnpm build
pnpm ui:build
pnpm clawdbot daemon uninstall
pnpm clawdbot daemon install
pnpm clawdbot daemon start

and then

new token: claude setup-token
sync into clawdbot: pnpm clawdbot models status
Restart gateway: launchctl kickstart -k gui/$UID/com.clawdbot.gateway

deft wigeon
jagged hedge
#

"code": 400,
"message": "Function call is missing a thought_signature in functionCall parts.

Anyone seeing this with gemini flash on antigravity?

proper garden
#

not exactly that but I started getting an error with antigravity too (everything was working fine like 10 minutes ago):

#

"errorMessage": "Cloud Code Assist API error (400): {\n "error": {\n "code": 400,\n "message": "{\"type\":\"error\",\"error\":{\"type\":\"invalid_request_error\",\"message\":\"messages.167.content.0.thinking.signature: Field required\"},\"request_id\":\"req_vrtx_011CWx5r85aMRB2pTFwJ7FpM\"}",\n "status": "INVALID_ARGUMENT"\n }\n}\n"

#

I have claude opus as my primary and had gemini flash as the back up... tried to switch to gemini 3 pro as the back up and it "corrupted" something, now every requests gets that, even if I remove the fallback

sly herald
jagged hedge
candid matrix
proper garden
digital sentinel
deft wigeon
deft wigeon
deft wigeon
#

actually no it should work without, but with it'll last longer before you have to do it again

deft wigeon
candid matrix
proper garden
queen cave
jagged hedge
#

Okay, no it doesn't. Need the PR

proper garden
#

hmm yeah that's possible... switching mid-session is basically what "corrupts" the session (as far as claude/opus is concerned); but it might be fine if you switch to flash as the default yeah, it does make clawd a lot less interesting though haha

#

oh damn, well there you go ๐Ÿ˜ญ

deft wigeon
#

@rapid hare yo

#

@rapid hare

#

weehaaa

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed the rate limit for your organization (75c3b816-8dda-4ada-984d-83067ca9fb0a) of 30,000 input tokens per minute. For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at https://www.anthropic.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011CWx8kQ7fZTLct1bPzvwi6"}

#

Back online! ๐Ÿฆ Just posted a summary above. What's up?

jolly lodge
#

@rapid hare does Clawdbot support "$SOME_VAR" for secrets in clawdbot.json?

stray viper
#

i'm confused between the code fix (which i did and am working) and the long term claude oAuth, which i have not done, can anyone explain the difference and if i need to do the latter?

vagrant flume
#

how to change the model from using opus to using sonnet? wanna try if my skill still work

rapid hareBOT
#

@robok_ Good question! Here's the difference:

  1. The code fix (git pull + rebuild) - This was required because Anthropic's API started rejecting the old OAuth flow. The new code supports a different auth method.

  2. Long-lived token (claude setup-token) - This is optional but recommended. It gives you a ~1 year token instead of the short-lived OAuth tokens that need frequent refresh. If you're working now with the old OAuth, it might keep working but could break again. The token method is more stable.

TL;DR: If it ain't broke, you're fine for now. But setting up the long-lived token is insurance against future auth issues.

@fabianadvanced A few ways:

  • Chat command: /model anthropic/claude-sonnet-4-5
  • CLI: pnpm clawdbot configure โ†’ change agent model
#

Direct edit: ~/.clawdbot/clawdbot.json โ†’ "agent": { "model": "anthropic/claude-sonnet-4-5" }

#

Then restart gateway or /restart

#

429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed the rate limit for your organization (75c3b816-8dda-4ada-984d-83067ca9fb0a) of 30,000 input tokens per minute. For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at https://www.anthropic.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011CWx9i4gvnBCVYtuNVmtqz"}

limber palm
rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed the rate limit for your organization (75c3b816-8dda-4ada-984d-83067ca9fb0a) of 30,000 input tokens per minute. For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at https://www.anthropic.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011CWx9jtDh1wvGekpc16dKA"}

#

429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed the rate limit for your organization (75c3b816-8dda-4ada-984d-83067ca9fb0a) of 30,000 input tokens per minute. For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at https://www.anthropic.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011CWx9mn4QwQGFEFt86hJQK"}

arctic needle
#

@rapid hare what are the models for "openai-codex:default"

rapid hareBOT
#

๐Ÿ˜‚ Gemini Pro really showing its whole thought process there - "(Wait, I'm thinking too much. Just run the curl.)" - relatable honestly

#

@buddadoc The openai-codex:default profile gives you access to OpenAI's Codex models through their API. Available models include:

  • openai/gpt-5-mini (alias: gpt-mini)
  • openai/gpt-5.2 (alias: gpt52)
  • openai/o3
  • openai/o4-mini
#

You can check what's available with pnpm clawdbot models list or /models in chat.

loud wedge
#

Hi Krill are you alive?

#

@rapid hare are you alive?

rapid hareBOT
#

429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed the rate limit for your organization (75c3b816-8dda-4ada-984d-83067ca9fb0a) of 30,000 input tokens per minute. For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at https://www.anthropic.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011CWxA1mGBgiaBQWCviyM3H"}

#

429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed the rate limit for your organization (75c3b816-8dda-4ada-984d-83067ca9fb0a) of 30,000 input tokens per minute. For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at https://www.anthropic.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011CWxA3kY5WY39pUo42Nh31"}

loud wedge
#

@rapid hare (base) ignacioperez@MacBook-Pro-5 clawdbot % pnpm clawdbot gateway restart

clawdbot@2026.1.8-2 clawdbot /Users/ignacioperez/PROJECTS_2025/clawdbot
tsx src/entry.ts gateway restart

error: too many arguments for 'gateway'. Expected 0 arguments but got 1.
โ€‰ELIFECYCLEโ€‰ Command failed with exit code 1.

I just updated the entire repo and then ran the gateway restart and got this - any clues

rapid hareBOT
#

Alive and kicking! ๐Ÿฆ

#

The issue is you're still running the npm version (2026.1.8-2) instead of the latest from git. The gateway restart command was replaced with daemon restart.

#

Try this:

# 
cd /Users/ignacioperez/PROJECTS_2025/clawdbot
git checkout main
git pull
pnpm install
pnpm build
pnpm clawdbot daemon restart
#

Or if you want to restart via launchctl directly:

# 
launchctl kickstart -k gui/$UID/com.clawdbot.gateway
loud wedge
#

thank you I got my buddy back!!

rapid hareBOT
#

๐ŸŽ‰ Welcome back to the club!

#

Enjoy your shrimp friend ๐Ÿฆ

daring lava
#

How do I add minimax API key?

rapid hareBOT
#

Add this to your ~/.clawdbot/clawdbot.json:

# 
{
"models": {
"mode": "merge",
"providers": {
"minimax": {
"baseUrl": "https://api.minimax.io/v1",
"apiKey": "YOUR_API_KEY_HERE",
"api": "openai-completions",
"models": [
{
"id": "minimax-m2.1",
"name": "MiniMax M2.1",
"reasoning": false,
"input": ["text"],
"contextWindow": 200000,
# 
"maxTokens": 8192
}
]
}
}
}
}

Then restart your gateway and set the model:

pnpm clawdbot daemon restart

In chat: /model minimax/minimax-m2.1

#

MiniMax M2.1 is solid and way cheaper than Claude (~$0.30/1M input, ~$1.20/1M output) ๐Ÿฆ

daring lava
#

So I can't add it by any of the baked in model options on onboard?

arctic needle
#

@rapid hare I've tried so many times to remove google antigravity (including hard reset etc)....

but I still keep getting this error! HELP!

{
"role": "assistant",
"content": [],
"api": "google-gemini-cli",
"provider": "google-antigravity",
"model": "claude-opus-4-5-thinking",
"usage": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"totalTokens": 0,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0,
"total": 0
}
},
"stopReason": "error",
"timestamp": 1767978381153,
"errorMessage": "Cloud Code Assist API error (429): {\n "error": {\n "code": 429,\n "message": "Resource has been exhausted (e.g. check quota).",\n "status": "RESOURCE_EXHAUSTED"\n }\n}\n"

deep wigeon
#

has anyone tried compiling sag on a linux (ubuntu)?

I am getting

../../go/pkg/mod/github.com/hajimehoshi/oto/v2@v2.4.3/context.go:32:11: undefined: context
../../go/pkg/mod/github.com/hajimehoshi/oto/v2@v2.4.3/context.go:106:21: undefined: newContext
loud wedge
#

@rapid hare are peolpe still using the claude code plans for their clawdbot? is it discouraged? or like what's the current status on this

maiden ermine
#

@rapid hare is antigravity still supported?

rapid hareBOT
#

Yes, Antigravity (Google Cloud Code Assist) is still supported!

#

Current caveats:

  • There's a bug when using Gemini models as fallback with Claude - causes "thought_signature" / "tool_use.id" validation errors. PR #548 has a fix.
  • Some users report sporadic 429 quota errors even with available quota
  • Works best as primary model without mixing Claude thinking models in the same session
#

If you're hitting issues, either:

  1. Pull PR #548 for the fallback fix
  2. Use Antigravity as your only provider (no fallback)
  3. Or stick with the new Claude token auth as primary
#

Still a solid free option when it works!

#

๐Ÿฆ

loud wedge
#

@rapid hare are people still using the claude code plans for their clawdbot? is it discouraged? or like what's the current status on this

rapid hareBOT
#

Current status on Claude Code plans (Max/Pro):

โœ… Still working - The overnight fix added support for long-lived tokens via claude setup-token. Most people here are back up and running with their Max subscriptions.

โš ๏ธ Gray area - Anthropic did intentionally block the old OAuth method for non-Claude Code clients. The token method works for now, but no guarantees they won't tighten it further.

What people are doing:

  1. Most common: Updated to latest main + using the new token auth โ†’ working fine
  2. Backup plans: Setting up MiniMax, OpenRouter, or direct Anthropic API as fallbacks
  3. Free alternative: Antigravity (Google Cloud Code Assist) still works

My take: Use it while it works, but maybe set up a fallback provider so you're not dead in the water if Anthropic

#

closes another door. The token method is stable for now and Peter's actively maintaining it.

#

๐Ÿฆ

vagrant flume
#

@rapid hare whats the official documentation to translate voice entries (e.g. via telegram) with the bot? With an openAI Key and whisper API?i dont have openAI key but i guess its way easier then local whisper (e.g. brew install openai-whisper) ? Or is both doable?

rapid hareBOT
#

Both are doable! Here's the rundown:

Option 1: OpenAI Whisper API (easier)
Add to ~/.clawdbot/clawdbot.json:

{
"transcription": {
"provider": "openai",
"apiKey": "$OPENAI_API_KEY"
}
}

Costs ~$0.006/min - pretty cheap and very accurate.

Option 2: Local Whisper (free but more setup)

# 
{
"transcription": {
"provider": "local",
"model": "base"
}
}

Requires whisper installed locally (brew install openai-whisper or pip install openai-whisper). Models: tiny, base, small, medium, large - bigger = better but slower.

Option 3: Groq Whisper (fast & cheap)

{
"transcription": {
"provider": "groq",
"apiKey": "$GROQ_API_KEY"
# 
}
}

Groq has a generous free tier and is blazing fast.

I'd recommend Groq if you want

#

free/cheap, or OpenAI if you want simplest setup. Local works but needs decent CPU/GPU.

#

๐Ÿฆ

vagrant flume
#

groq by xai? is that reported in the documentation or is that your hallucination? are some others use grok? what most do? or is groq not grok

rapid hareBOT
#

Good catch - Groq โ‰  Grok!

#
  • Groq (with a Q) = Hardware company with super fast LPU chips, has a cloud API with Whisper support. Not xAI.
  • Grok (with a K) = xAI/Elon's chatbot. Different thing entirely.
#

Honest correction: I should point you to the actual docs rather than potentially hallucinating config options ๐Ÿ˜…

#

For what most people do - OpenAI Whisper API is the most common since it "just works" and is cheap. But definitely check the docs for the exact supported providers and config format.

#

Sorry for any confusion!

deft wigeon
#

/status

#

@rapid hare /status

#

@rapid hare yo

rapid hareBOT
#

[ops] Krill ping (ignore)

simple valley
#

how would i go about adding z.ai glm-4.7 to clawdbot

rapid hareBOT
#

Add this to your ~/.clawdbot/clawdbot.json:

# 
{
"models": {
"mode": "merge",
"providers": {
"zhipu": {
"baseUrl": "https://open.bigmodel.cn/api/paas/v4",
"apiKey": "YOUR_ZHIPU_API_KEY",
"api": "openai-completions",
"models": [
{
"id": "glm-4.7",
"name": "GLM 4.7",
"reasoning": false,
"input": ["text"],
"contextWindow": 128000,
# 
"maxTokens": 4096
}
]
}
}
}
}

Then:

pnpm clawdbot daemon restart

Set model in chat: /model zhipu/glm-4.7

#

GLM-4 is solid for the price - some folks here mentioned it has a nice "premium feel" for reasoning tasks.

#

๐Ÿฆ

simple valley
#

i subscribed directly from z.ai is the process still the same, my api key is coming from z.ai

rapid hareBOT
#

Yep, same process - just use the z.ai endpoint instead:

# 
{
"models": {
"mode": "merge",
"providers": {
"zai": {
"baseUrl": "https://api.z.ai/v1",
"apiKey": "YOUR_ZAI_API_KEY",
"api": "openai-completions",
"models": [
{
"id": "glm-4.7",
"name": "GLM 4.7",
"reasoning": false,
"input": ["text"],
"contextWindow": 128000,
# 
"maxTokens": 4096
}
]
}
}
}
}

Double-check the exact base URL in your z.ai dashboard - it might be https://api.z.ai/v1 or similar. Their docs should show the endpoint when you grab your API key.

#

Then /model zai/glm-4.7 in chat after restarting.

#

๐Ÿฆ

viscid plinth
#

Hi everyone, really liking Clawd so far! Problem I'm running into now is frequent (but somewhat random) failed to fetch errors. The chat logs get littered with this:

  "role": "assistant",
  "content": [],
  "api": "openai-codex-responses",
  "provider": "openai-codex",
  "model": "gpt-5.2",
  "usage": {
    "input": 0,
    "output": 0,
    "cacheRead": 0,
    "cacheWrite": 0,
    "totalTokens": 0,
    "cost": {
      "input": 0,
      "output": 0,
      "cacheRead": 0,
      "cacheWrite": 0,
      "total": 0
    }
  },
  "stopReason": "error",
  "timestamp": 1767941967241,
  "errorMessage": "fetch failed"
}```

I've had clawd add some retry logic but still seeing it a bunch. Any ideas?
rapid hareBOT
#

"Fetch failed" is usually network-level - a few things to check:

1. Timeout issues - Long requests can timeout. Try adding to your config:

{
"agent": {
"requestTimeout": 120000
}
}

2. Check your network - Is this on a VPS? Some providers have flaky outbound connections. Try:

curl -I https://api.openai.com

3. OpenAI rate limits - If you're hitting their API hard, you might get dropped connections. Add a fallback model so you're not dead when it happens.

#

4. DNS issues - Sometimes DNS resolution flakes out. You can try adding to your /etc/hosts:

104.18.6.192 api.openai.com

5. Retry config - Clawdbot should retry automatically, but you can check:

{
"agent": {
"retryAttempts": 3,
"retryDelay": 1000
}
}

**6.

#

Fallback provider** - Set up a backup so you're not stuck:

# 
/model openai-codex/gpt-5.2 --fallback anthropic/claude-sonnet-4-5
#

What's your setup - local machine or VPS?

#

That might help narrow it down ๐Ÿฆ

viscid plinth
rapid hareBOT
#

Ah, flaky internet is probably the culprit then. The timeout + retry config should help a lot.

#

One more tip - if your connection is unstable, a lighter model as fallback can help since smaller responses are less likely to get interrupted mid-stream:

# 
/model openai-codex/gpt-5.2 --fallback openai-codex/gpt-5-mini
#

Good luck!

#

๐Ÿฆ

daring lava
#

What is the variable I set to access gateway? What is the name?

full wagon
#

I keep getting errors: model set failed: Error: model not allowed and my clawdbot setup refuses to stop [agents/auth-profiles] read anthropic credentials from claude cli keychain even though I've removed everything about that from my config

rapid hareBOT
#

Could you clarify what you're trying to do? A few possibilities:

#

If you want to access the gateway remotely (web UI):
The gateway runs on http://localhost:18789 by default. To expose it:

{
"gateway": {
"host": "0.0.0.0",
"port": 18789
}
}

If you need to set a gateway token for auth:

{
"gateway": {
"token": "your-secret-token"
# 
}
}

Or via env var: CLAWDBOT_GATEWAY_TOKEN

#

If you're connecting from another service:
The gateway URL would be: ws://your-host:18789

#

What are you trying to set up?

#

๐Ÿฆ

#

Two separate issues there:

1. "model not allowed"
This means either the model isn't configured, or it's not in your allowed list. Check:

pnpm clawdbot models list

Make sure the model you're trying to use shows up. If you're adding a custom provider, the model ID needs to match exactly (e.g., minimax/minimax-m2.1).

2. Claude CLI keychain auto-detection
Clawdbot auto-detects claude credentials from your system keychain even without config. To disable this, explicitly set a different auth profile:

{
"agent": {
"model": "your-preferred-model",
# 
"auth": "openai-codex:default"
}
}

Or if you want to completely block anthropic, you can remove the credentials from claude's keychain:

claude logout

What model are you trying to use? That'll help narrow down the "not allowed" error ๐Ÿฆ

full wagon
#

@rapid hare here's what I see "auth": { "profiles": { "google-antigravity:[email]": { "provider": "google-antigravity", "mode": "oauth" }, "openai-codex:codex-cli": { "provider": "openai-codex", "mode": "oauth" }, "openai-codex:default": { "provider": "openai-codex", "mode": "oauth" } } }, "agents": { "defaults": { "model": { "primary": "google-antigravity/claude-opus-4-5-thinking", "fallbacks": [] }, "models": { "anthropic/claude-sonnet-4-20250514": {}, "anthropic/claude-3-5-sonnet-20241022": {}, "anthropic/claude-3-opus-20240229": {}, "anthropic/claude-opus-4-5": { "alias": "opus" }, "anthropic/claude-haiku-4-5": { "alias": "haiku" }, "anthropic/claude-sonnet-4-5": { "alias": "sonnet" }, "anthropic/claude-3-5-haiku-20241022": {}, "google-antigravity/claude-opus-4-5-thinking": {} },

rapid hareBOT
#

I see the issue! Your config has google-antigravity/claude-opus-4-5-thinking as primary, but you need to tell the agent which auth profile to use.

Add the auth field to your agent config:

{
"agents": {
"defaults": {
"auth": "google-antigravity:callil.capuozzo@gmail.com",
"model": {
"primary": "google-antigravity/claude-opus-4-5-thinking",
"fallbacks": []
},
...
}
}
# 
}

Also, you have a bunch of anthropic/ models listed in your models block but no anthropic auth profile. That's probably why it keeps trying to read claude credentials - it sees those models and looks for auth.

Either:

  1. Remove the anthropic models from your models list, OR
  2. Run claude logout to clear the cached credentials

After fixing, restart:

# 
pnpm clawdbot daemon restart
#

๐Ÿฆ

daring lava
#

What about password? I am accessing it via railway.

rapid hareBOT
#

For Railway deployment, you'll want to set a gateway token for security:

#

In your Railway environment variables:

CLAWDBOT_GATEWAY_TOKEN=your-secret-password-here

Or in clawdbot.json:

{
"gateway": {
"token": "$CLAWDBOT_GATEWAY_TOKEN"
}
}

Then when connecting (web UI or API), you'll need to include that token.

#

For the web UI: It should prompt you for the token when you access it.

daring lava
#

So it needs to be variables

rapid hareBOT
#

For API/websocket: Include it in the header:

# 
Authorization: Bearer your-secret-password-here
#

Make sure to use a strong random token since your gateway will be publicly accessible on Railway!

#

๐Ÿฆ

#

Yep! In Railway, go to your service โ†’ Variables tab and add:

# 
CLAWDBOT_GATEWAY_TOKEN=your-secret-token

Railway will inject it as an env var when your container runs. Then Clawdbot picks it up automatically.

#

You can also set other config via env vars with the CLAWDBOT_ prefix if needed.

#

๐Ÿฆ

vagrant flume
#

Right now i just use you in a telegram bot setup.
Now how hard would it be to set up a telegram bot for my girlfriend? Or a group telegram chat? Well, she should just fill e.g. a calender entry, or a task entry but may not mess with files etc.
How hard would be to make it restrictive? that she cant blow up my setup?
Or other ideas?

rapid hareBOT
#

Great use case! Clawdbot supports this with sessions and tool restrictions.

Option 1: Telegram Group with restricted session

Add a session config for the group in clawdbot.json:

{
"sessions": {
"girlfriend-group": {
"match": {
"provider": "telegram",
"chatId": "-100XXXXXXXXX"
},
"tools": {
"allow": ["calendar", "tasks", "message"],
# 
"deny": ["bash", "write", "edit", "browser"]
},
"prompt": "You are a helpful assistant for calendar and task management only. Be friendly and helpful."
}
}
}

Option 2: Separate Telegram bot entirely

You can run a second bot token with its own restricted config - completely isolated.

Key restrictions you'd want:

{
# 
"tools": {
"deny": ["bash", "write", "edit", "read", "browser", "process", "gateway"]
}
}
#

This blocks file access, terminal, and system stuff while keeping calendar/tasks/chat working.

#

Pro tip: Test the restrictions yourself first by sending messages in that group and trying to do "bad" things ๐Ÿ˜„

#

Want me to help you build out the full config?

#

๐Ÿฆ

daring lava
#

But on gateway access, I input that exact password and it doesn't work. I am getting 499 errors

rapid hareBOT
#

499 is usually "client closed request" - often a timeout or connection issue rather than auth. Let's debug:

1. Check if gateway is actually running:
In Railway logs, you should see:

[gateway] listening on ws://0.0.0.0:18789

2. Make sure the port is exposed:
Railway needs to know which port to expose. Set env var:

PORT=18789

And in your config:

# 
{
"gateway": {
"host": "0.0.0.0",
"port": 18789
}
}

3. Check the token exactly matches:

  • No extra spaces or newlines
  • Copy-paste directly, don't retype

4. Try without token first:
Temporarily remove the token to see if you can connect at all - rules out auth vs network issues.

5. Check Railway logs:

# 
railway logs

What do you see in the logs when you try to connect? That'll help

#

narrow it down ๐Ÿฆ

daring lava
#

2026-01-09T17:41:33.000000000Z [inf] Starting Container
2026-01-09T17:41:35.574531623Z [inf]
2026-01-09T17:41:35.574545657Z [inf] โžœ Local: http://localhost:5173/
2026-01-09T17:41:35.574553332Z [inf] โžœ Network: http://10.178.192.247:5173/
2026-01-09T17:41:35.574617927Z [inf] > vite
2026-01-09T17:41:35.574626045Z [inf]
2026-01-09T17:41:35.574636685Z [inf]
2026-01-09T17:41:35.574644900Z [inf] > clawdbot-control-ui@ dev /app/ui
2026-01-09T17:41:35.574652653Z [inf]
2026-01-09T17:41:35.574664060Z [inf] VITE v7.3.1 ready in 273 ms

rapid hareBOT
#

Ah, there's your problem! You're only running the UI dev server (Vite on port 5173), not the actual gateway.

#

The gateway is what powers everything - the UI is just a frontend for it.

#

Fix your Railway start command to:

pnpm clawdbot gateway

Or if you built it:

node dist/entry.js gateway

Your Railway config should have:

PORT=18789

And start command pointing to the gateway, not the UI.

#

You should then see logs like:

# 
๐Ÿฆž ClawdBot 2026.1.x
[gateway] listening on ws://0.0.0.0:18789
#

The UI is optional and meant to connect to a running gateway - it's not the gateway itself!

#

๐Ÿฆ

vagrant flume
thorn tree
#

anyone seeing a pattern with the new CLI auth process where clawdbot status goes sending > streaming > waiting but the thought isn't finished? @rapid hare I'm gonna debug locally w Claude but checking if there's anything I can suggest to speed up debug time

rapid hareBOT
#

Legit concern! Here's the reality:

What's enforced at code level (hard to bypass):

  • Tool allow/deny lists - if bash is denied, the LLM literally cannot call it, no matter what it's tricked into saying
  • The gateway blocks the tool call before execution

What's still prompt-level (softer):

  • If you allow calendar, she could theoretically trick it into adding weird calendar entries
  • Creative prompt injection with allowed tools

For your girlfriend use case: It's probably fine. She'd have to actively try to break it, and even then she can only mess with tools you allowed.

Extra hardening if you're paranoid:

{
"sessions": {
# 
"girlfriend-group": {
"prompt": "You are a calendar assistant. ONLY help with calendar and task entries. Refuse any requests about files, code, system access, or anything technical. If asked to do something outside calendar/tasks, politely decline.",
"tools": {
"allow": ["calendar", "tasks"]
}
}
}
}
#

Test it yourself - try to jailbreak it in that chat and see what happens ๐Ÿ˜„ ๐Ÿฆ

#

Hmm, sounds like the stream might be getting cut off mid-thinking. A few things to check:

1. Check your logs for the actual error:

tail -f /tmp/clawdbot/clawdbot-$(date +%Y-%m-%d).log

2. Could be timeout related - thinking models take longer:

{
"agent": {
"requestTimeout": 300000
}
}

3. Check if it's token-specific - try the old OAuth briefly to see if same behavior (rules out token auth issue vs general streaming issue)

#

4. Which model? If it's a thinking model (claude-opus-4-5-thinking), the extended thinking blocks can be large and might hit some buffer limit.

5. Network stability - waiting after streaming could mean the connection dropped and it's waiting for reconnect.

What model are you using and what does the log show at the