Gmail already sorts by date. InboxSift sorts by what the message asks of you. That is a model plus guardrails, not a magic unread badge.
Category vs type
Category is Act, Know, or Noise: how much attention you should spend. Type is newsletter, receipt, travel, school/work, social, and so on. A receipt can be Know. A newsletter can be Noise. A flight change can be Act and still be type travel.
What the classifier sees
For recent Inbox threads, InboxSift uses sender, subject, snippet, a clipped body preview, and bulk-mail headers (List-Unsubscribe, List-Id). That content goes to our classify API so a model can return a category and Briefing line. We do not pretend the model “never reads” the email.
Your overrides
Always Act / Know / Noise on a contact is a hard preference. Relationship hints (important vs automated) are a weak prior when the model is unsure—not a way to hide a real homework assignment.
Typical Act / Know / Noise examples
- Act: “Please confirm Tuesday at 3,” homework assigned, bill to pay, permission request.
- Know: Apple receipt, package delivered, meeting room changed, grade posted.
- Noise: product roundups, “jobs picked for you,” kudos, class enrollment ads.
Where this runs
The Chrome extension calls /api/inbox-sift/inbox and /api/inbox-sift/classify. Results cache locally so Briefing feels instant on reopen. Full HTML loads per thread when you open it.
Frequently asked questions
- Can classification be wrong?
- Yes. Override the contact, or open the thread in Gmail. Briefing is a filter, not a court.
- Does Noise mean deleted?
- No. Noise is omitted from Briefing. The message stays in Gmail.
Sources & further reading
Published 2026-09-04. See also our full article library.