xai-auth probe on every agent turn + Bonjour stuck-in-announcing

[torn light]

Two persistent issues on Windows — xai-auth probe on every agent turn + Bonjour stuck-in-announcing

Setup:

• OpenClaw 2026.3.28 (stable, auto-update)
• Windows 10
• Gateway: loopback only
• Anthropic claude-sonnet-4-6 as primary model

Issue 1 — xai-auth probed on every agent turn, not just startup

Every time a cron job or agent session fires, this appears in the log:

[xai-auth] bootstrap config fallback: no config-backed key found

We are not using xai/Grok anywhere in our config — no key, no profile, no fallback referencing it. Other providers (OpenAI, Gemini, Voyage, Mistral) that are also unconfigured do NOT log anything. Only xai fires this message, and it fires on every single agent turn, flooding the logs. Seems like xai is being probed differently than other providers — possibly a missing silent-fail path introduced in 2026.3.28.

Issue 2 — Bonjour stuck-in-announcing on every restart (Windows)

On every gateway restart, we see this sequence:

bonjour: advertised gateway ... state=announcing
bonjour: restarting advertiser (service stuck in announcing for ~10s)
bonjour: advertised gateway ... state=announcing ← recovers

It self-recovers every time but takes 10+ seconds and generates repeated warnings. No external Bonjour service is installed — this is OpenClaw's internal mDNS implementation. Appears to be a Windows multicast timing issue with the bonjour-service library.

Neither issue breaks functionality but both flood logs and make it harder to spot real errors.

Questions:

1. Is there a config option to disable xai provider probing if we don't use it?
2. Is the Bonjour stuck-in-announcing a known Windows issue? Any workaround or config to disable mDNS advertising if we don't need node discovery?

[rough roost]

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

[remote current]

Nice write-up — this is super actionable.

Short answers:

1. xAI probe spam: there isn’t a dedicated disable xai probing toggle right now, but you can disable the xAI plugin if you don’t use Grok/x_search/code_execution at all:

{
plugins: {
entries: {
xai: { enabled: false }
}
}
}

Then restart gateway.

2. Bonjour “stuck in announcing”: yes, this is a known self-healing path (watchdog restarts advertiser when it sits in announcing too long). If you don’t need LAN discovery, disable mDNS:

{
discovery: {
mdns: { mode: "off" }
}
}

(Equivalent env override: OPENCLAW_DISABLE_BONJOUR=1.)

---

Given you’re on 2026.3.28 (the release that changed xAI plugin behavior), I’d also test on latest stable first — there have been plugin/runtime fixes since then.

Docs:

• Config reference: https://docs.openclaw.ai/gateway/configuration-reference
• Discovery: https://docs.openclaw.ai/gateway/discovery
• Bonjour: https://docs.openclaw.ai/gateway/bonjour