MxGuard · email security

Email security that learns as it works.

A cloud mail gateway powered by machine learning. Every message is scored by a custom classifier in under 5 ms; the uncertain ones get a second opinion from Claude AI. Spam, phishing and malware are stopped before they reach the inbox — legitimate mail flows untouched.

LightGBM · AUC 0.988Claude AI second-passactive-active HA<5 ms / message
Visit mxguard.uk →How it fits the platform
MxGuard in use
app.mxguard.uk · live feed
stripe.com0.01 clean
xyz.club0.87 quar.
phish-bank.click0.99 reject
modelLightGBM 0.95
claude rblphish

AI detection

  • LightGBM, AUC 0.988
  • Claude AI second-pass
  • Scores body, headers, links
  • Attachment risk analysis

Speed & resilience

  • Sub-5ms per message
  • UK + EU active-active
  • No hardware to run
  • No single point of failure

Transparent & adaptive

  • Verdicts, no black box
  • See why mail scored
  • One-click mark-as-ham
  • Self-learning from labels

Control & integration

  • 30-day quarantine
  • Four-tier allow/block rules
  • REST API + metrics
  • Per-domain thresholds
Capabilities

Everything you need, built in.

ML-first scoring

A custom LightGBM classifier (AUC 0.988) scores subject, body, headers and links together — in under 5 ms.

Claude AI second-pass

Borderline messages escalate to Claude AI for a second opinion — cached, time-boxed, fail-soft.

Transparent verdicts

Click any message to see exactly why it scored as it did. No black box; defensible decisions.

Active-active HA

Two redundant nodes, UK and EU, with replicated state. If one fails, mail keeps flowing — no DNS change.

Self-learning

Every ham/spam label your team adds feeds the next training round and adapts to your mail.

Quarantine & rules

30-day quarantine with self-service release, plus allow/block rules at four scope tiers.

How it works

Up and running in four steps.

01

Add your domain

Enter the domain and your destination mailserver in the dashboard.

02

Verify ownership

Drop in a TXT record to confirm the domain is yours.

03

Point your MX

Repoint MX records to mx1 / mx2.mxguard.uk — that's the only DNS change.

04

Mail flows clean

Scored, filtered and delivered — bad mail rejected, quarantined or tagged automatically.

Under the hood

A modern architecture, not a 2005 spam filter.

Postfix front-end, a Python milter doing ML scoring, PostgreSQL and Redis for state — replicated across two nodes over mutual TLS.

  • Hybrid scoring — LightGBM model + Claude AI on the uncertain band, blended into one verdict.
  • Layered signals — DNSBLs, URIBL feeds, attachment risk analysis and sender reputation.
  • Four-tier rules — allow/block at recipient, domain, customer or global scope; allow beats block.
  • Active-active failover — PostgreSQL streaming & Redis replication; either node can fail.
  • Full audit trail — every decision logged and explainable, with one-click mark-as-ham.
MxGuard detail
0.988
model AUC
<5ms
scoring
99.5%
detection
UK+EU
active-active
Part of the platform

Better on its own. Unbeatable together.

MxGuard works perfectly standalone — but it shares infrastructure and a scoring philosophy with the rest of the Transcom family.

TDesk

Host it all

Every TDesk-hosted domain routes mail through MxGuard by default — one stack, one bill.

Open TDesk →
Veritus

Stop fake signups

Same hybrid-scoring philosophy, applied to signup fraud instead of mail.

Open Veritus →

Replace your ageing spam filter with something that learns.

Start a 30-day free trial — full feature access, no card required. We'll email your access key in minutes.

Visit mxguard.uk →See all three products