Nine signals first. AI only for the few.
Blockjay sits between your mail provider and you: it reads each incoming message, decides whether it's spam, and sorts it — spam to a private quarantine, everything else left in your inbox. You review and correct the results in one app, and every correction makes the filter more yours.
Cheap, fast checks first — the expensive one last.
Every message runs the same funnel and the filter stops the moment a layer is confident. The early layers are free; the AI step is a genuine last resort.
Your allow / block rules
Senders you've pinned are an instant decision — no further checks needed.
Authentication
SPF, DKIM and DMARC check the message really came from where it claims.
Reputation & lists
Known-bad senders, domains and IPs are caught outright, before they reach you.
Fuzzy hashing
Near-duplicate spam campaigns are recognised from a single seen copy.
Your personal Bayes
Statistics trained only on your corrections — not a shared, one-size model.
Blockjay AI — last resort
Only the genuinely borderline middle is escalated — to our own self-hosted model, with Anthropic as a fallback — and only a short excerpt is ever sent, never your whole mailbox.
Connect once, then review and correct.
Connect the mail you already have
Connect Gmail or Microsoft with a single sign-in, or any mailbox over IMAP or POP3. No new inbox, no migration, no change to your address — Blockjay layers on top. For IMAP it watches and acts on your inbox directly; for POP3 it keeps full copies so nothing is lost. Credentials are encrypted and never logged.
Run the funnel — cheap checks first
Each message passes through the layers below and stops as soon as one is confident. Clearly valid mail is delivered untouched; clearly spam is moved to quarantine. Only the uncertain few reach the AI — with only the minimum (sender, headers, a short excerpt), never your whole mailbox, under a zero-retention, no-training arrangement. You can turn AI escalation off entirely.
See why
Nothing is a black box. Each message carries a score and a one-line reason, and the detail view lists the contributing signals. Spam and phishing are rendered defensively — remote images off, links defanged, scripts stripped — so you can look without risk.
Your corrections train your filter
When you restore a message (“not spam”) or flag a missed one (“mark spam”), Blockjay moves it, updates your allow/block list, and teaches your Bayesian filter. Because that training is per-user, your filter fits how you use email — not an average of a billion strangers. A week or two of light correcting and it settles into a quiet, accurate rhythm.
Watch it work before you let it act.
New mailboxes start in monitor mode: Blockjay classifies everything but moves nothing, so you see exactly what it would catch. When you're convinced, flip to enforce mode and it starts moving spam out of your inbox. Either way, nothing is ever deleted.