Hebrew & Arabic Text Toolkit: Normalize, Number to Words, Gematria

Fast batch utilities for Hebrew and Arabic text that general-purpose tools get wrong. Built for developers, no-code automations (Make, Zapier, n8n) and AI agents via MCP.

Operation	Example input	Example output
normalize – for search, deduplication, RAG and embeddings	בְּרֵאשִׁית בָּרָא / السَّلَامُ عَلَيْكُمْ ١٢٣	בראשית ברא / السلام عليكم 123
numberToWords – amount in words for invoices, receipts and cheques	1234.50 (ILS)	אלף מאתיים שלושים וארבעה שקלים וחמישים אגורות
numberToWords (Arabic)	2.25 (SAR)	اثنان ريالان و خمس و عشرون هللة
hebrewNumeral – Hebrew-letter numbers for dates, chapters, lists	5787	תשפ״ז
gematria – standard, gadol and katan	שלום	376
fixKeyboardLayout – text typed in the wrong layout	akuo / יקךךם	שלום / hello
detect – script, direction, niqqud/tashkeel presence	שלום עולם hello	{"script":"mixed","direction":"rtl",…}

Why

• Search & RAG: vowelled (niqqud) Hebrew and harakat Arabic don't match unvowelled queries; Arabic letter variants (أ/إ/ا, ى/ي, ة/ه), tatweel (ـ), Arabic-Indic digits and presentation-form ligatures from PDFs break exact matching and embeddings. normalize fixes all of these with switchable options.
• Finance & invoicing: Israeli and Arab-market documents often require the amount in words. Hebrew grammar (gender agreement: שקל אחד, שני שקלים, אגורה אחת, שתי אגורות) is handled correctly.
• Judaica, calendars, education: Hebrew numerals with geresh/gershayim (ט״ו, ט״ז special cases) and gematria.
• Support & forms: repair text typed with the wrong keyboard layout (Israeli SI-1452 ↔ US QWERTY).

Input example

{
  "operation": "numberToWords",
  "numberMode": "currency",
  "currency": "ILS",
  "texts": ["1", "2", "1234.50", "0.02", "₪ 1,000"]
}

Output example

[
  { "index": 0, "operation": "numberToWords", "input": "1", "result": "שקל אחד", "status": "ok" },
  { "index": 1, "operation": "numberToWords", "input": "2", "result": "שני שקלים", "status": "ok" },
  { "index": 2, "operation": "numberToWords", "input": "1234.50", "result": "אלף מאתיים שלושים וארבעה שקלים וחמישים אגורות", "status": "ok" },
  { "index": 3, "operation": "numberToWords", "input": "0.02", "result": "שתי אגורות", "status": "ok" }
]

Invalid items return "status": "error" with a message and are not charged.

Options

• normalize: removeDiacritics, unifyArabicLetters, replaceHebrewFinalLetters, asciiDigits, asciiPunctuation
• numberToWords: numberLanguage (he/ar), numberMode (cardinal/ordinal/currency), currency (Hebrew: ILS, USD, EUR, GBP; Arabic: SAR, EGP, KWD, TND), hebrewGender
• gematria: gematriaMethod (standard/gadol/katan)
• hebrewNumeral: omitThousands (5787 → תשפ״ז or ה׳תשפ״ז)
• fixKeyboardLayout: layoutDirection (auto / en->he / he->en)

Pricing

$0.001 per text processed ($1 per 1,000). Errors are free. Up to 100,000 texts per run.

Use from AI agents (MCP)

Add the Actor to the Apify MCP server and your agent (Claude, Cursor, ChatGPT connectors, LangGraph…) can call it as a tool: "write 1,234.50 shekels in Hebrew words".

Notes

• Hebrew number words follow standard modern Hebrew; review legal documents as usual.
• Arabic number words come from the open-source num2words library (LGPL-2.1, used unmodified).
• No data leaves the run; texts are not logged.