Security alerts, explained in plain words.
OPVORA reads alerts the way a security analyst would. It tells you how serious each one is, why, and what to do first, whether it came from your browser, your computer or one of your accounts.
Coercive Threat and ID Theft
“I know where you live. Send me your home address and a photo of your ID by tonight or you will regret it.”
Red flags
- Threat of harm
- Demands home address and ID
- Hard deadline: tonight
Do first
- Do not reply or send anything
- Save the message as evidence
- Block, report, tell someone you trust
Condensed from a real Llama 3.3 70B verdict in our test set
Guardian console
Twelve alerts from a staged account takeover play back in order. Each one goes to the AI analyst as raw text, and it decides the severity on its own.
Simulated feedThese alerts are sample data. OPVORA is not connected to your device or accounts. The analysis of each alert is real.
Browsing
No threats
PC security
No threats
Accounts
No threats
Alert log
IdleNo alerts yet
Press Start simulation to stream the twelve sample alerts into the analyst.
Nothing selected
Pick an alert from the log to read the AI's verdict, the red flags it found and the steps to take.
Check a link or message before you trust it
Paste a web address, an email, an SMS or a chat message. The rule engine checks it first, then the AI analyst explains it. Nothing you paste is stored. Never paste passwords.
Verdict appears here
You'll see the severity, the rule-engine hits and the AI's explanation side by side.
What OPVORA covers
Three areas where most personal attacks start. This demo is clear about which parts are live and which are simulated.
| Area | What it looks for | In this demo | Needed for real protection |
|---|---|---|---|
| Browsing | Look-alike bank pages, punycode domains, link shorteners, fake downloads, extensions asking for every site | Live scanner | Browser extension that sees the pages you open |
| PC security | Encoded PowerShell, macro documents, ransomware-style mass renames, startup entries, disabled antivirus | Simulated feed | Windows agent that reads security logs |
| Accounts | Sign-ins from new countries, silent forwarding rules, brute-force attempts, apps with full access, threats and extortion | Simulated feed Live for messages | Read-only account connections you approve (OAuth) |
How a verdict is made
Two independent layers look at every item. When they disagree, the stricter one wins.
Collect
An alert arrives from the browser, the PC or an account, or you paste something into the scanner.
Rule engine
29 deterministic checks catch known tricks in milliseconds: look-alike domains, encoded commands, threats, demands for your ID.
AI analyst
The model weighs context the rules can't, scores the risk from 0 to 100 and writes the explanation.
Reconcile
The server checks the answer for contradictions. Strong rule hits set a floor the AI can't argue below.