Inbox: 9,748 → 5,277

Request: „Take care of my inbox.“

1. Apply the rule table: first matching rule wins, deterministic, no language model involved.
2. Lift receipts into the document archive with checksum verification, deadlines into the calendar, people facts into the contact dossier.
3. Move the rest into a deletion pipeline with a four-week grace period.
4. Derive an inventory of all online accounts from the senders — 573 services.

Almost ten thousand e-mails in the inbox, receipts scattered among them, deadlines buried in subject lines.

Final deletion, every send, activating new rules.

A doctor's reply with alternative appointments sat unnoticed in the spam folder for 18 days — a script now checks spam monthly.

A rule table where the first matching row wins is the opposite of a chatbot: traceable, testable, delegable.

“Take care of my inbox” is a single sentence, and behind it runs a rule table where the first matching row wins. Receipts move into the document archive with checksum verification, deadlines into the calendar, the rest into a deletion pipeline with a four-week grace period. Deleting and sending stay with me. The backlog fell from 9,748 to 5,277 e-mails, on the way to inbox zero; an inventory of 573 online accounts emerged as a by-product. The hard part was not the AI, it was writing down the rules.