Live metrics across all processed mailboxes.
The latest mail moved to Junk. Tap Diagnostics for the full breakdown.
Attachments checked against VirusTotal by hash.
Classification tokens the AI has used. The monthly estimate is based on a TYPICAL day (median), so testing bursts like backfills don't inflate it. Billing is in Cloudflare Neurons; this is a rough estimate.
Plus the flat $5/mo Workers Paid plan. The estimate uses only automatic runs, so your testing doesn't inflate it, and shows "collecting…" until a couple of real days pass. Open Cloudflare dashboard for exact Neuron usage.
The mailboxes this filter processes. Add them one at a time; click the × on a chip to remove it. Every mailbox in the list is active. An empty list means ALL mailboxes in the tenant.
Mailboxes here are never processed, even if the list above is empty (all). Exclude wins.
Check the sender's real sending domain against DNS domain blacklists. A listed domain is marked Junk immediately, without asking the AI.
Block or allow specific senders. Enter a full email address (blocks that sender) or a bare domain (blocks the whole domain and its subdomains). Blocked = auto-Junk. Allowed senders are never junked and always classified normally. Allow wins over block.
The AI model that reads each email and picks its category. Larger models categorize more accurately; the light one is faster.
Every email is scored from clues (authentication, links, content, blacklists, sender history). This slider is the score at which mail is treated as spam. Lower = more aggressive. Mail near the line gets an AI second opinion.
Labels the AI applies to kept mail. Pinned ones are always offered; it can also learn new ones as it sees new kinds of mail.
Neutralize links in junk so they can't be clicked, while keeping the real URL readable.
An invisible marker we set on every message we touch, so the same mail is never processed twice. It does not appear as a category or tag in your mailbox.
Prepend the category to every message's subject (e.g. "[Newsletter] ...") so it's visible in any mail app and usable in mail rules. Note: this changes the subject, so it will appear on replies/forwards.
The last 30 emails scored, with the exact clues that fired. This is private to you and never written onto the mail. Use it to see why something scored the way it did. Turn on AI reasoning while troubleshooting to have the AI explain its verdict (auto-off after 1 hour to save tokens).
Result of the most recent processing pass.
—
Community reputation built from mail moves. When someone rescues mail from Junk that sender gains reputation (green); when someone junks mail it loses reputation (red). Recent votes count most, so a domain that goes bad drops fast. This nudges the spam score but never overrides threat scanning.