When does the AI automatically switch a conversation to Human mode?

HostPal flips to Human mode automatically in three specific situations. In every other case, the AI keeps replying — even during emergencies.

1. You replied manually — sending a message in the inbox auto-pauses the AI so it doesn't double up. Each reply you send resets a 2-hour timer; after 2 hours of no host activity, the conversation auto-reverts to AI.

2. The bot detected a complaint — if the guest's message reads as a real complaint ("this place is filthy", "the WiFi has been broken for two days", "I want a refund"), the AI tags its own reply as needing human follow-up and pauses. You'll see "complaint" as the reason in the conversation sidebar.

3. The bot looped on low confidence — if the AI produced three unhelpful, hedging replies in a row (typically because the guest is asking about something not in your knowledge base), it gives up and hands off to you. An incident is automatically created.

4. A restricted topic was hit — if you've configured "restricted topics" for a property (e.g. "refunds", "subletting"), any guest message matching those terms locks the conversation immediately.

What does NOT auto-lock:

• Life-safety emergencies (gas leak, fire) — the bot stays in AI and replies with safety instructions immediately while you get paged via Telegram/WhatsApp. Locking would delay help.
• Prompt injection or jailbreak attempts ("ignore previous instructions", "DAN mode") — bot refuses politely and stays in AI. Locking on every curious guest tester would be hostile.
• Abusive language without a complaint — bot refuses and stays in AI.

The intent: keep the bot answering routine questions, hand off cleanly when something needs you, and never leave a guest hanging.