Visibility and Granular Control for AI Auto-Reply Learning and Modes

Problem Statement:
The AI currently learns from past inbox messages ("auto-detected" data), but hosts have no visibility into what has been learned or the ability to edit/delete this data. This leads to the AI occasionally surfacing outdated or incorrect third-party information (e.g., old rental car contacts) in guest-facing replies. Additionally, Auto Reply is an all-or-nothing setting, meaning hosts cannot distinguish between low-risk factual topics and high-risk recommendation topics.

Proposed Solution:
1. AI Learning Management: Provide a way for hosts to view, edit, and delete the "auto-detected" information the AI has learned from past inbox messages.
2. Granular Auto-Reply Modes: Implement the ability to set Auto Reply to 'Active' for low-risk, factual topics (e.g., WiFi, check-in/out) while automatically switching to 'Draft-only' (Suggest with AI) for topics involving third-party recommendations or high-consequence information.

Unlock:
This would allow property managers to maintain high automation for routine queries while ensuring that any potentially sensitive or outdated information is reviewed by a human before reaching the guest. This is not a hypothetical risk: there have been documented cases where Auto Reply sent guests specific third-party phone numbers and email addresses that were never verified or approved, pulled from old inbox messages that can no longer be traced or corrected. For property managers running dozens of units, this creates ongoing liability exposure.