Sentry demo

Sentry · Email-threat triage

Submit an email. Get a verdict in seconds.

Sentry fans an inbound message out to four specialist analyzers in parallel, fuses their signals into a single verdict, and returns an LLM-generated natural-language explanation of why it ruled the way it did. Every path you see in this demo hits the same orchestrator and broker that production clients use — no mocked results.

What runs behind the scenes

Header analyzer
SPF / DKIM / DMARC alignment plus heuristic checks on forensic headers (Received chains, Message-ID spoofing).
URL scanner
Threat-intel lookups (VirusTotal, Google Web Risk, PhishTank) and pattern checks against every embedded link.
Domain reputation
RDAP registration data, age heuristics, and domain intel against the sender's envelope-from domain.
Attachment scanner
File-type + macro + hash-reputation checks on every attachment. Returns red on VT hits or risky MIME types.

The orchestrator waits for the tool fan-in, runs verdict fusion, asks the configured LLM (Azure OpenAI in canary) for a human-readable explanation, and writes the result to the sentry/pipeline-event MQTT topic so downstream consumers (this demo page, the admin portal, audit stores) receive the same payload.