When you need to take the scan system down for maintenance, a new toggle in /admin/app-settings stops every new scan — whether triggered from the UI, the API, MCP, or autopilot — while anything already running finishes cleanly. Users who try to scan while it's paused see a plain modal you can customize, and the API returns a clear error rather than silently failing. In-flight scans complete normally.
The hosting wizard now handles the Cloudflare case that quietly breaks agent feeds: when a bot filter is blocking the verification path, the wizard names the problem and explains why it matters. That same filter blocks ChatGPT, Perplexity, and Claude from reading your feed — which means the feed can't get you recommended even when it's otherwise working. The wizard walks you through a path-scoped WAF Skip rule and flags that Cloudflare's "Block AI bots" setting overrides per-bot allowlists, so the exemption has to target your feed path directly. If your feed is genuinely live but the automated check can't get through, you can now set the subpath as your primary destination without waiting on a verification your bot filter won't allow.
The chat agent is more capable in two ways: you can now paste a URL directly into a conversation and it reads that page rather than searching around it, and it handles real math — growth rates, totals, projections — rather than estimating. Several reliability problems are also resolved. The agent no longer hangs when a scan it started fails; it reacts immediately, tells you what happened, and offers to retry. Long-running scans and reopened threads both pick up results correctly now instead of timing out or stalling.
The website-matched header now loads your brand's actual typeface instead of falling back to a generic serif, and dropdown menus in cloned headers no longer jump to a corner of the screen on hover. Notification toasts cap at four visible at a time and always drain on their own — a burst of background events no longer builds into a permanent stack.