Live call
- Real mediaclear
- Real cameraclear
- Live personclear
A real person.
Really there. Really human. On calls, selfies, and signups.
Trust, checked.
Is it Trustable?
Anyone can be faked. Anything can be forged.
One API checks the person, the media, and the paperwork.
Before money moves.
Policy
—
Person
Checking…
45+kinds of fake, covered
~240 msmedian verdict, on live video
SOC 2 Type IIReality Inc. · Report and DPA on request
3 waysto run it: cloud, on-prem, on-device
The story
He saw his CFO. He heard his colleagues. He wired $25 million. No one else on the call was real.
Hong Kong, 2024. Nothing was hacked. He trusted what he saw. In 2025 the FBI logged 22,364 AI-enabled fraud complaints and $893M in losses. Source: FBI IC3 ↗
Looks real.
AI-generatedIt isn't.
A face used to be proof. So did a voice. So did a document. Now they're just files.
Every loan, claim, and wire now rests on them.
Trust Check
One API. All evidence.
Image · video · audio
- AI-generatedflagged
- Editedclear
- Cloned voiceclear
Real media.
Made by a camera, or by a model.
Bank statement
- Pixel forensicsflagged
- Fonts & templateclear
- Cross-file patternflagged
Real paperwork.
The numbers they sent, or the ones they edited.
AudioPartner
Voice, through our partner Modulate. Same API, same verdict.
ComplianceNew
AI-made media, labeled as it should be.
AI agentNext
Who runs it. Who it answers to.
Trustable SealIn build
Checked once. Trusted everywhere.
Evidence
Don't take our word. Take the evidence.
Every verdict comes with a confidence score and the signals behind it. Ready for your auditor.
Evidence recordSample
- Check
- chk_8f2c91
- Subject
- person · ses_8f2c91
- Verdict
- Trustable · 0.97
- Signals
- real media clear
real camera clear
live person clear - Model
- detect-1.6
- Media kept
- none (default)
- Request
- req_41d07c
Why Trustable
One answer. Not a piece of one.
Detectors check media. Identity vendors check IDs. Fraud sends both, plus a PDF.
| Trustable | Deepfake detectors | Identity verification | |
|---|---|---|---|
| Question | Is this real, right now? | Is this file fake? | Who is this person? |
| Covers | People, images, video, audio, documents. Agents next. | Media | People, at checkpoints |
| On a fail | Verdict, signals, record | A score | Pass or fail |
| Documents beyond IDs | Yes | No | Rarely |
| Live + uploaded | Both | Media, no context | At checkpoints |
Use cases
Start where money moves.
KYC · Lending · Fintech
Before the account opens.
Synthetic applicants. Injected selfies. Forged papers. Flagged before approval.
Insurance claims · SIU
Before the claim is paid.
Photos, video, paperwork. Verified, with evidence you can defend.
Finance & treasury
Before the wire goes out.
Check the approver. And everyone else on the call.
Platforms · Agents · Next
Before the agent gets access.
Know who the agent answers to.
Under the hood
Not a promise. In production.
Detection model
Video and image. 45+ kinds of fake.
On-device engine
Live calls, checked where they happen.
Device signals
A real camera, or a feed pretending to be one.
Document forensics
IDs, statements, invoices, claims.
API and on-prem
One endpoint. Your cloud, or ours.
Risk analytics
Attack patterns across your applications and claims.
The API
One call. One verdict.
Sync. JSON in, verdict out. Send a file by URL, or a live session from the SDK. Retry safely with an idempotency key.
# Before the wire goes out POST /v1/checks Authorization: Bearer tk_live_… Idempotency-Key: wire_7f3a91 { "subject": "person", "source": { "session": "ses_8f2c91" } } # 200 · sync · ~240 ms, live video { "id": "chk_8f2c91", "verdict": "trustable", "confidence": 0.97, "signals": { "real_media": "clear", "real_camera": "clear", "live_person": "clear" }, "model": "detect-1.6", "media_retained": false, "request_id": "req_41d07c" } # Files: "source": { "url": "…" } # trustable | review | not_trustable
Privacy & security
Prove it's real. Reveal nothing else.
- Trust, not identityWe tell you it's real. Not who it is.
- On-device firstLive video stays on the device when it can.
- Analyze, then forgetMedia is discarded after analysis, by default. The evidence record stays.
- Show the workEvery verdict lists its signals.
CloudOn-premOn-device
Imagine
Imagine a world where every voice is a person. Every photo is a moment. Every document says what it says. That's the internet worth rebuilding.
Checked once. Carried everywhere. No name attached. That's the Seal.
- Applicants who exist.
- Claims that happened.
- Statements nobody edited.
- Wires that reach the right hands.
Questions buyers ask
Straight answers.
Isn't detection an arms race?
Against one signal, yes. So we never rely on one.
How accurate is it?
Don't trust a number on a website. We'll run it on your files and show you.
We already have an IDV vendor.
Keep them. They tell you who. We tell you if it's real.
What happens to our media?
Analyzed, then discarded. On-device or on-prem if you need it.
What about false positives?
You set the threshold. Grey cases go to review, evidence attached.
How long to integrate?
One REST call for files. An SDK for live calls. Beside your IDV, not instead.
Pricing?
Per check, with volume tiers.
Banks · Insurers · Platforms
Make it Trustable.
- Send 20 files. Some you know are fake.
- Verdicts back before the call.
- Building with agents? Design-partner seats are open.