Manifesto

Safety that understands.

AI that sees, hears and reads what people share, and stops harm while it happens.

Use cases

Where moderation can’t wait.

  • 12.4K
  • 318
  • 94
@lina.movesnew routine ✨ #fyp

this is so good 🔥

how old r u? dm me, don’t tell ur parents 👀

Grooming risk · hidden
Child safety

Strangers reach children in comments and DMs.

Risky messages hidden before a child reads them.

squad-voice4 in voice

“tell anyone and I’ll find you”

Threat · 00:42
Voice chat

In games, abuse is spoken, not typed.

Voice checked live, down to the second.

Blurred on air
LIVE
1.2M watching
Live broadcast

Live TV has no second take.

Every frame checked as it airs, blurred on the spot.

Review queue
18,432
items waiting · ~30 s each
Detectivision · 0.2 s each
Real-time

People can’t keep up. Reviewing is slow and exhausting.

AI clears the queue in moments; people make the hard calls.

47:31 / 1:12:40
47:31 · hate speech
Late night talk · Ep. 42
1:04:12
2:11:09
Long videos

One frame looks harmless. The whole hour isn’t.

Every minute checked, with a timeline of what was found.

#game-club

anyone wanna play later?

me! 🎮

ur so mature for 12. send ur address, i have a gift 🎁

Hidden · sent to safety team

my number is 0532 ••• •• ••

Personal data masked
Communities

Children overshare and strangers ask for more.

Personal data masked, strangers’ requests stopped.

Photos: Tan Kaninthanond, ActionVance, l ch, Cody Board, Fábio Magalhães, Nainoa Shizuru / Unsplash

Our vision

Moderation should understand, not just filter.

  1. Real time.

    Decisions in milliseconds, inside your product.

  2. Every format.

    Text, images, video, audio and live streams in one API.

  3. Context.

    Meaning comes from the whole, not a single frame.

  4. People decide.

    AI finds it fast; people make the hard calls.

  5. Private by design.

    Our own models, nothing kept longer than needed.

  6. Measured openly.

    We publish what we catch and what we miss.

Languages

Every language deserves its own AI.

Translations of English are not enough. Each language needs investment in its own models, built on real data with real R&D.

  • Türkiye

    Turkish, spoken by more than 80 million people, with models of its own.

  • Europe

    Not only the big languages: we aim to cover Hungarian, Bulgarian, Albanian and every smaller one.

  • Arabic

    We aim for Arabic and its dialects too, across the Gulf and beyond.

Let’s build it together.

— The Detectivision team · 2026