Telegram shows billing error while Web UI responses succeed (Claude Max)

[afiffattouh]

Summary

What went wrong?
I use Claude Max subscription. Limits are still not hit. OpenClaw response is normal in the WebUI. Telegram receives this error:
"⚠️ API provider returned a billing error — your API key has run out of credits or has an insufficient balance. Check your provider's billing dashboard and top up or switch to a different API key."

Steps to reproduce

1. I send a prompt through Telegram.
2. Telegram bot is alive "Typing".
3. I receive "⚠️ API provider returned a billing error — your API key has run out of credits or has an insufficient balance. Check your provider's billing dashboard and top up or switch to a different API key."
4. I check WebUI, the response is being generated and I can read OpenClaw messages.
5. I use Claude Max sub (not hitting the limits).

Expected behavior

The responses that are being generated in the WebUI should be delivered to the Telegram bot

Actual behavior

Instead, I receive "⚠️ API provider returned a billing error — your API key has run out of credits or has an insufficient balance. Check your provider's billing dashboard and top up or switch to a different API key."

Environment

• Clawdbot version: latest 2026.2.6
• OS: Ubuntu
• Install method (pnpm/npx/docker/etc): docker

Logs or screenshots

Paste relevant logs or add screenshots (redact secrets).

[lukaslau]

I am also getting same error. Channel: Slack, Model: Kimi k2.5, API key has enough credit.

[jabezborja]

Getting same error.

[afiffattouh]

this is OpenClaw diagnosis after numerous times trying to understand the issue (not sure if it's accurate but worth highlighting):

OpenClaw applies sanitizeUserFacingText() to all messages delivered to Telegram (but NOT to WebUI's websocket stream)
That function checks if the response text matches billing error patterns:
Contains /\b402\b/ (the word "402")
Contains "credit balance"
Contains "billing" + "credits" or "payment" or "plan"
We've been talking about pricing, billing, credits, API errors all day — my responses naturally contain these words
OpenClaw thinks my response IS a billing error and replaces my entire response with the ⚠️ billing error message
WebUI works because it receives raw WebSocket stream events (no sanitization). Telegram fails because outbound delivery runs through sanitizeUserFacingText() which false-matches my response content.

This is an OpenClaw bug — the sanitizer shouldn't match on normal response content, only on actual error messages.

[afiffattouh]

[afiffattouh]

We have all of them from the source code:

Pattern Group 1 (any match triggers):

• 402 (as a standalone number)
• "insufficient credits"
• "credit balance"
• "plans & billing" (exact phrase)

Pattern Group 2 (combination triggers):
Must contain "billing" AND one of:

• "upgrade"
• "credits"
• "payment"
• "plan"

Safe words (won't trigger alone):

• "billing" by itself — safe
• "credits" by itself — safe
• "payment" by itself — safe
• "plan" by itself — safe
• "upgrade" by itself — safe

Examples that WOULD trigger:

• "Check your billing plan" ❌
• "Billing and payment options" ❌
• "Your credit balance is low" ❌
• "Error 402" ❌

Examples that are SAFE:

• "The project plan looks good" ✓
• "I'll credit you for that" ✓
• "Send the payment tomorrow" ✓
• "Monthly subscription" ✓

That's the complete list from the source.

[afiffattouh]

There are actually 2 call sites, not 1:
┌──────┬────────────────────────────┬─────────────────────────────────────────────────────────┬─────────────────────────┐
│ Line │ Function │ Purpose │ Should patch? │
├──────┼────────────────────────────┼─────────────────────────────────────────────────────────┼─────────────────────────┤
│ 523 │ formatAssistantErrorText() │ Formats actual error messages (stopReason === "error") │ No - this is legitimate │
├──────┼────────────────────────────┼─────────────────────────────────────────────────────────┼─────────────────────────┤
│ 535 │ sanitizeUserFacingText() │ Runs on all outgoing text including normal AI responses │ Yes - this is the bug │
└──────┴────────────────────────────┴─────────────────────────────────────────────────────────┴─────────────────────────┘
Assessment: The patch is safe and correct. Here's why:

1. Line 535 is the only problematic one. sanitizeUserFacingText() runs on every message sent to users. When the AI talks about "billing plan" or
   "credit balance", the entire response gets nuked and replaced with the billing error banner.
2. Line 523 is fine to leave. formatAssistantErrorText() only runs when stopReason === "error" (line 505: if (msg.stopReason !== "error" && !raw)
   return). This correctly catches real billing errors from the API.
3. isBillingAssistantError() (line 622-624) is also fine. It explicitly checks stopReason === "error" before calling isBillingErrorMessage().
4. Risks are minimal:
   - Only side effect: if the API returns a billing error as the message content (not as an error), it won't get caught. This is extremely unlikely -
   billing errors come as error responses, not content.
   - Gets overwritten on npm update - you'd need to re-apply.

Recommendation: Patch only line 535 in all 3 files. Leave line 523 alone.

[afiffattouh]

I applied the patch and it worked:

[lukaslau]

I applied the patch and it worked:

Can you check if it works correctly in case you get "insufficient balance" error from API and add credit to API? It might be the issue because my first error was real. I added more credit and was getting same error but the next day (or maybe after VPS restart) error was gone.

[afiffattouh]

It's a message filtration function issue. even if you have sufficient credit, and any of those words are mentioned in the bot response you will receive the error: #11359 (comment)

I applied the patch directly into the code and now it's working fine. Also someone already fixed it and submitted a PR for merging. Next release it should be fixed.