Biometric
Document OCR · Document Verification

433+ document types. 1.2 seconds. Less manual review.

The customer photographs the document — the system extracts the data, checks the MRZ, finds signs of retouching and confirms authenticity. One API call instead of manual moderation.

Try for freeOpen the documentation
433+document types
1.2 secdata extraction
99%+OCR accuracy
MRZvisual checks

Clients who trust us

NurbankOtbasy BankBinancenaimi.kzBank RBKTennisi.kzSun FinanceShinhan FinanceKcellNPCK — National Payment Corporation of Kazakhstan
Authenticity check

We cut off 99.9% of forged and edited documents

The core checks not just the text but portrait geometry, fonts and the microstructure of characters, MRZ checksums and signs of re-capture — in the same 1.2 seconds.

01

Photo substitution

Detects a pasted-in or re-shot portrait: a mismatch of photo edges, lighting and face geometry against the document zone.

02

Retouching and data tampering

Pixel analysis reveals altered fields: font anomalies, cloning traces and broken character alignment.

03

Screen & paper copy

We reject a document re-shot from a monitor or printout: moiré, glare and the pixel grid give away the re-capture.

04

Expired and invalid

We check the validity period, number format and MRZ checksums — a document with a mismatched check digit is rejected automatically.

Documents & templates

180+ countries. 12,000 templates. One API call.

The system detects the document type itself and breaks it down into structured fields. 180+ countries and 12,000+ document templates — with no manual tuning for each format.

UX + Analytics

Smart frame capture — the customer shoots the document on the first try

The system doesn't just return a “Pass / Fail” verdict. It helps the customer complete the check on the first try and collects metadata for your security and compliance team.

Document-frame hints
The screen prompts — “Fit all corners in the frame”, “Remove the glare”, “Hold the document flat”. Fewer retries and less drop-off.
Shot quality control
We check sharpness, glare, tilt and frame completeness before submission — so OCR gets a readable document on the first try.
Metadata for compliance
Document type, MRZ status and authenticity result stay in the session record for your security team.
On-Premise deployment

Passport data never leaves your loop

Run in our cloud, or deploy the recognition core into a closed perimeter on your own servers (On-Premise) — scan copies and personal data stay with you.

The result: document images never leave your company perimeter. On-Premise helps you meet local storage and data-control requirements; what applies depends on the jurisdiction and your internal rules.

ISO 27001GDPRNational Bank requirements
Cloud · SaaSOn-Premise
Your company's perimeter
OCR core
Scan copies
Data never leaves the perimeter
For developers

Integration in a few hours

A ready camera widget, native SDKs or a direct REST API — recognition and authenticity check in a single response.

JS · Webview

Web SDK / Flow Widget

A ready JS camera widget with auto frame capture for a website or embedded browser (Webview). Customizable to your brand book.

JavaScriptAuto-captureWebview
iOS · Android

Mobile SDK

Native libraries with direct camera access and local document cropping — they work even on a weak connection.

SwiftKotlinOffline-ready
Backend

REST API

Direct access to protected document-recognition endpoints for your server-side logic. Authorization via a private X-API-KEY.

RESTX-API-KEYServer-side
AI · AGENTS

MCP server for AI agents

AI agents connect to document recognition directly — no code required. Compatible with Claude, GPT and any MCP client.

MCPClaudeAI-agents
document — REST API
# Document recognition and check
POST /v1/documents/recognize/
Host: api.biometric.vision
X-API-KEY: sk_live_••••••••••••••
Content-Type: multipart/form-data
# → 200 OK
{ "doc_type": "passport", "country": "RUS",
"fields": { "surname": "Ivanov", "name": "Ivan",
"birth_date": "1985-06-12", "number": "1203567890" },
"authentic": true, "mrz_valid": true, "score": 0.998 }
API response

What document recognition returns

The response is JSON: the extracted fields, the document type that was recognised and the result of the technical checks. Below are the key fields and the errors worth handling on your side.

Key fields

FieldWhat it means
first_name, last_name, date_of_birth, document_numberThe fields extracted from the document.
date_of_expiryExpiry date — used to check the document is still valid.
document_type.document_nameThe recognised type, e.g. “Kazakhstan — Id Card (2014–2022)”.
document_type.fdsid_list.d_mrzWhether this side carries an MRZ line for checksum validation.
document_type.check_authenticityThe authenticity checks performed for the recognised template.
images.face_photoThe portrait from the document — the input for face matching.
failure_reasonWhy the document was rejected, when it was.

Common errors

ErrorWhat to do
Frontside document not foundThe front side did not read: ask the user to retake the shot.
Document image at frontside_image field is not frontsideThe two sides were swapped in the request.
Not allowed file, valid extensions: .jpg .png .jpegUnsupported file format — convert it before sending.
Client does not have access to Document Recognition TechnologyNo active subscription for the technology, or access is switched off.

The current list of recognised document types and the full response format live in the documentation. See the list of document types

Numbers & facts

433+
Document types —Support for passports and hundreds of other document types worldwide
180+
Countries supported —Global coverage for checks across different jurisdictions
40 sec
Onboarding time —User signup and verification in under a minute
99%
Technology accuracy —High-precision AI-based document recognition and verification
FAQ

Questions about document verification

Recognition (OCR) extracts the fields: name, date of birth, number and expiry. The authenticity check works on the image itself — MRZ checksums, fonts, the geometry of the photo area and traces of editing. They are separate checks and come back separately in the response.

The document type is detected automatically; you do not pick a template. The current list of recognised types is published in the documentation — check it there rather than against a marketing page.

The capture widget checks corners, glare and tilt before the image is sent and asks the user to retake it. If the shot still fails, the API returns an error — “frontside document not found”, for example — and you decide whether to retry or send the application to manual review.

Yes. Besides the cloud, an On-Premise deployment is available: the recognition core runs inside your perimeter and scans never leave it.

No. A document check confirms the document, not that the person presenting it owns it. That is why face matching against the document portrait and a liveness check sit next to it.

Next in the flow: match the document holder · confirm a live person is present