Help — Welcome to MxGuard

MxGuard is an AI-powered email security gateway built from the ground up with modern infrastructure and machine learning at its core.

What it does

MxGuard sits in front of your mail server as your MX record. Every inbound message is scored against a trained ML model plus multiple complementary signals (URL reputation, IP reputation, header heuristics, threat feeds). Mail that scores clean is forwarded to your mail server unchanged. Mail that looks like spam is held for review. Mail that's clearly malicious is rejected at SMTP time.

Why it's different

Traditional anti-spam stacks (SpamAssassin, Rspamd) rely on rule trees built up over decades. They work, but the rules are noisy, hard to tune, and slow to adapt. MxGuard takes a different approach:

One ML model trained on your actual mail corpus, scoring messages 0–1
Customer self-learning — every time you mark something spam or ham, the system remembers and applies that signal to future mail from the same sender
Sub-5ms scoring per message — no Bayesian database to maintain, no per-rule lookups
Real-time live feed — see every scan as it happens, with full score breakdown
Modern stack — PostgreSQL, Redis, Prometheus, FastAPI, all running on Alma Linux 8

Why it's better

Higher accuracy on the spam you actually see — the model is trained on real in-flow mail, not generic public datasets
Self-service domain management — customers can add and verify their own domains via DNS TXT records
Transparent scoring — for every message you can see exactly why it scored as it did (model score, URL reputation hits, IP reputation, heuristics)
Lower false positives — legitimate bulk-sender VERP envelopes (Shopify, Mailchimp, Currensea, etc.) are recognised structurally and not penalised
Per-domain customisation — different thresholds per domain, tag mode for customers who prefer delivery-with-warning over hard quarantine
Daily digest emails — see what was held with one-click release/spam links

Where to next

Use the sidebar to navigate. If you're new, start with Quick start. For an overview of how the scoring pipeline works, see How scoring works.