Pause before you tap.

Check a message for common scam patterns. Get a clear reason and a safer next step in English or Arabic.

Go to the message checker

Your message stays in this browser. Standard Check works offline; AI mode runs locally.

A phone message held in an open hand beneath three protective signal waves
Check a message together.See what the tool noticed, not who sent it.
Choose a check

AI mode downloads its runtime and about 268 MB of model weights from jsDelivr and Hugging Face. Your message is analyzed in this browser and is not uploaded.

Remove names, phone numbers, and account details first.

0 / 1,600

All examples are invented for practice.

Ready when you are

Nothing checked yet

Paste a message to see what this tool notices. It cannot confirm who sent it.

This tool can miss scams or flag genuine messages. It cannot confirm who sent a message. Verify unexpected requests using a contact method you already trust.

A model we can question.

The text model learns patterns from labeled practice messages. This panel checks it against examples it did not train on.

—Loading practice-set results

Held-out synthetic examples
LanguageClass accuracyFalse positivesFalse negativesUncertain

Four-class accuracy measures the model label. A binary alert is “high caution” or “pause and verify”; uncertain results count as no alert and are included in false negatives for scam examples. These invented examples are not real-world accuracy.

What happens to the text?

Three separate steps keep the result understandable.

  1. 1

    Learned language patterns

    A small Naive Bayes classifier compares words and character patterns with labeled English and Arabic examples.

  2. 2

    Visible warning signs

    Separate checks look for requests for codes or money, urgency, and links. The app shows which clues it found.

  3. 3

    Safer choices

    The result suggests how to verify independently. It never opens a link or declares a sender authentic.

A practice tool, not a security service.

This prototype uses synthetic training data and can miss scams or flag ordinary messages. Standard Check works offline. AI-Assisted Check downloads a local runtime and about 268 MB of model files on first use; the message text stays in this browser and is not sent to a cloud AI service. Do not enter private information. If a message asks for a password, card details, or a one-time code, do not share them; verify through the organization’s official app or a contact method you already trust.