Documents in, JSON out

OCR for teudat zehut, passports, driver's licenses and bank cheques — Hebrew and Latin fields, check-digit validation, public registry checks, one API call.

Request accessBy invitation · 5 documents free Get itSelf-host from GitHub

CHEQUE · BANK 11148

Pay to: —
Amount: 4,500.00
Date: 2026-09-07

⑈80001234⑈ 11 14841 ⑆ 0000123456

תעודת זהות · ISRAELTD1

שם משפחה: ישראלי
שם פרטי: ישראל
מספר זהות: 1 2345678 2

I<ISR123456782<<<<<<<<<<<<<<<

PASSPORT · ISRP

Surname: ISRAELI
Given name: ISRAEL
Passport No.: 12345678
Date of birth: 01 JAN 1990

P<ISRISRAELI<<ISRAEL<<<<<<<<<<<<<<<<<<

passport_number12345678high

id_number123456782high

date_of_birth1990-01-01high

expiry_date2030-01-01medium

verified

Images are never stored
Check digits verified locally, no model in the loop
One multipart POST, JSON back

How it works

01Detect
Classical image analysis finds every document on the page — a card, a sefach, a cheque — and returns its box; the vision model is asked only when it finds nothing.
02Crop at full resolution
Each document is cut from the original image, never from a downscaled page, so small print stays readable.
03Extract
Structured output: the model fills a fixed schema — Hebrew names from the print, Latin names only from the MRZ, dates as ISO.
04Validate locally
ICAO 9303 MRZ and MICR check digits, the Israeli ID check digit and printed-vs-machine-read cross-checks — pure math, no model in the loop.

What is read

Six document types, one response shape.

Teudat zehut (ID card) + sefach
Both card layouts and the appendix sheet
- id_number
- first_name_he
- last_name_he
- date_of_birth
- date_of_issue
- date_of_expiry
- address
- marital_status
- spouse
- children[]
Israeli passport
TD3 MRZ with check digits
- passport_number
- mrz_lines
- last_name_en
- first_name_en
- last_name_he
- first_name_he
- nationality
- sex
- dates
Foreign passport
Any country, Latin names only
- passport_number
- mrz_lines
- last_name_en
- first_name_en
- nationality
- sex
- dates
Driver's license
Field numbers 4d, 5, 8, 9
- license_number
- id_number
- names
- dates
- address
- categories
Disability card
Names, ID and file number, validity month
- id_number
- file_number
- names
- date_of_expiry
Bank cheque
Front, and the guarantee stamp on the back
- micr_line
- bank
- branch
- account
- cheque_number
- drawer
- payee
- amount
- amount_words
- date
- guarantor

Public registries

Check what was read against Israeli public lists, or search them yourself.

Bank of Israel
Restricted bank accounts and severely restricted corporations
NBCTF
Designated terror operatives and organizations
Registrar of Companies
Registered companies, their status and the violator mark

On every extraction

Switch it on in settings or per request: every number and account read is looked up exactly, the full name as whole words.

check_registries=true
id_number
spouse_id_number
child_id_number
drawer_id_number
guarantor_id_number
account
name_he
name_en

Search

The Registries page and one API call search by number, bank account or name, results grouped by registry with the date of its data.

GET /api/v1/registries/search
id
bank
branch
account
name

Informational only: the lists are local copies as of their last download, a match may be wrong, and "not found" does not mean unlisted. Nothing read from a document is sent to any registry.

Privacy

Images are read in memory and discarded once the result is returned — never stored, never logged.
Per document we keep metadata — time, type, verdict, model, tokens, cost — never the image, its name or the values read from it.
The result is stored only if you switch it on, encrypted, until you delete it.
Nothing is sold or used for training. Servers are in Israel; documents are read by Anthropic's API and sign-in runs on Clerk, both in the US. The self-hosted variant with a local model never leaves your machine.

Personal data is processed under the Israeli Privacy Protection Law (Amendment 13).Read the Privacy Policy

Two ways to use it

Same engine, same API, your choice where it runs.

Cloud

Sign-up is by invitation: an admin invites you by e-mail, then 5 documents at our expense, then your own Anthropic API key — billed by Anthropic, no subscription here.

Request access

Self-host

Open source. One compose command brings up the web app, the engine and Caddy; Ollama runs the vision model on your own GPU.

AUTH_MODE=none MAKOR_BACKEND=ollama docker compose --profile ollama up -d --build

View on GitHub

API

Every account gets REST access with its own key. One multipart POST, JSON back.

curl -s -X POST https://makor.pro/api/v1/extract \
  -H "Authorization: Bearer ak_…" \
  -F "file=@document.jpg" \
  -F "check_registries=true"

response

{
  "document_type": "teudat_zehut",
  "fields": {
    "id_number": {
      "value": "123456782",
      "confidence": "high"
    },
    "last_name_he": {
      "value": "ישראלי",
      "confidence": "high"
    }
  },
  "validation": {
    "overall": "verified"
  },
  "registries": {
    "checked": [
      "id_number"
    ],
    "matches": []
  }
}

Read the API reference