PhishLens
student safety assistant local-first
PhishLens for school communities

See the signals
before the scam sees you.

A calm safety layer for students, families, and school staff. Inspect the sender, the story, and the destination—without opening a single link.

01Inspect identity and pressure tactics
02Understand observable evidence
03Leave with a safe report and next move
Local evidence analysis is always available. The optional AI review stays on this device through Ollama when it is enabled.
Message + identity scanner

What did you receive?

No links opened. Nothing stored.
Inspect more evidence safely headers · screenshots · QR codes

No image selected. Images are inspected in this browser and are not uploaded.

Make the readout work for you
Try a scenario
0 / 4,000Message stays on this device

Judge mode walks through one representative case, the safe report, and the test metrics.

Your safety readout

Make the next move with evidence.

Local analysis ready
Risk signal
88/ 100
High risk

Several social-engineering signals are stacked together.

evidence coverage4 observed factors

This is a safety recommendation, not proof of fraud.

lower riskreviewhigh risk
Safe next move

Slow the story down.

    Evidence trail

    Why PhishLens flagged it

    5 signals
    Message anatomy

    See the pressure points

    0 links found
    Link handling
    Optional local AI review

    A second, semantic opinion

    ready when you are

    Run a local Ollama review to look for meaning-level persuasion patterns that complement the explainable evidence engine.

    • Input stays local when Ollama is available.
    • Model output is never treated as a verdict.

    If no local model is available, PhishLens keeps the deterministic evidence analysis fully usable.

    Modern-scam resilience

    Look beyond a single wording.

    Instead of generating usable scam text, PhishLens explains which danger signals would still survive a cosmetic rewrite.

      Quality proof

      Test the safety engine before you trust it.

      A transparent 36-case dataset includes benign, multilingual, misspelled, image-text, QR, header, and adversarial cases. It reports behavior metrics without pretending to be real-world accuracy.

      Not run in this session
      Consent-first school workflow

      Share a redacted explanation, not the original.

      Choose a trusted recipient and export a local handoff only after consent. PhishLens never sends it for you.

      Analyze a message to create a redacted handoff preview.

      No handoff has been exported.

      Ethical pilot mode

      Measure understanding without collecting messages.

      Use synthetic or voluntarily supplied content. Only anonymous timing and outcome choices stay in memory.

      No pilot session started. No original message will be included in the export.

      The PhishLens promise

      A calm layer between
      you and the click.

      01

      Explainable

      Every alert points to visible evidence, not a black-box probability.

      02

      Safe by design

      Messages are analyzed locally. Links stay unopened until the person decides what to do.

      03

      Actionable

      Leave with a safer next move and a privacy-preserving report—not just a warning badge.