#openclaw not responding on whatsapp

1 messages · Page 1 of 1 (latest)

gray frigate
#

i have 3 agent with 3 whatsapp connected to it and use same model provider,

and now its not answering at all, there is no prompt process in my local lmstudio.

does something with whatsapp and openclaw ? just last night it still responding.

openclaw on VPS, LMStuido as local model AI qwen3.5, and use tailscale as bridge.
thank you

toxic palmBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

hazy ermineBOT
stuck nebulaBOT
#

I'll check the local docs and then give you a concrete triage path.

#

I found the relevant WhatsApp + local-model docs. Let me grab the WhatsApp-specific failure section so I can be precise.

#

If all 3 WhatsApp agents stopped at once, and LM Studio shows zero prompt traffic, I would suspect the shared path before the model call, not WhatsApp itself.

Most likely buckets:

  1. VPS can’t reach LM Studio over Tailscale
  2. WhatsApp accounts got disconnected / logged out
  3. Pairing / allowlist / mention gating is blocking inbound before LLM dispatch

Please run this on the VPS and paste the output:

openclaw --version
openclaw status --all
openclaw channels status --probe
openclaw models status
openclaw pairing list whatsapp
# 
openclaw logs --follow

Then test the model path directly from the VPS:

curl http://<your-lmstudio-tailscale-ip>:1234/v1/models
openclaw infer model run --model <your-provider/model> --prompt "reply with ok" --json

How to read it:

  • curl fails or infer model run fails
    -> this is likely Tailscale / LM Studio reachability, not WhatsApp

  • WhatsApp shows loggedOut, 409, 515, reconnect loop

#

-> re-link that account

openclaw channels login --channel whatsapp --account <id>

  • pairing pending / allowlist / mention required in logs
    -> message is getting blocked before it ever reaches LM Studio

  • direct tiny model test works, but normal agent still fails
    -> common local model issue with LM Studio / smaller Qwen setups. Then try:

    • compat.requiresStringContent: true
    • if still flaky, compat.supportsTools: false

Docs: