Thai Toolkit for AI Agents

One fast, offline Actor with ten Thai business and text tools that AI agents (via the Apify MCP server), LLM workflows, Make/Zapier/n8n automations and scripts can call over and over:

Operation	What it does	Thai keywords
bahtText	Amount in words for invoices and receipts, Thai or English	บาทถ้วน, ตัวอักษร, จำนวนเงินเป็นตัวหนังสือ
thaiDate	Parse Thai dates, convert Buddhist Era ↔ CE, Thai long/short formats, fiscal year	พ.ศ., ค.ศ., วันที่ไทย, ปีงบประมาณ
holidays	Bank of Thailand bank holidays and government holidays 2024–2027	วันหยุดธนาคาร, วันหยุดราชการ, วันหยุดชดเชย
businessDays	Is it a business day? Add N business days, count business days	วันทำการ
vat	VAT 7% add / extract with correct satang rounding	ภาษีมูลค่าเพิ่ม, VAT 7%
withholdingTax	Withholding tax by payment type (services 3%, rent 5%, advertising 2% ...)	ภาษีหัก ณ ที่จ่าย, ภ.ง.ด.3, ภ.ง.ด.53
taxIdCheck	Check digit and format of a 13-digit tax ID / company number	เลขประจำตัวผู้เสียภาษี, เลขทะเบียนนิติบุคคล
tokenize	Thai word segmentation	ตัดคำภาษาไทย
romanize	Thai → Latin letters (RTGS)	ถอดอักษรไทยเป็นโรมัน
address	Parse a Thai address, validate/fill the postcode, English names	แยกที่อยู่, รหัสไปรษณีย์, ตำบล อำเภอ จังหวัด

• Fast and cheap: no browser, no external calls, all data bundled in the image. Typical run is 2–3 s end to end at 256 MB; the work itself takes well under a second.
• Deterministic: the same input always gives the same output, so it is safe for agents to rely on.
• Batch friendly: send up to 10,000 items per run. Each item is processed on its own and returned in the same order.
• You pay only for items that succeed. Invalid items come back with an error and are not charged.

Input

{
  "operation": "bahtText",
  "items": [123.5, 21000000, "100.50"],
  "options": { "language": "th" }
}

• operation (required): one of the operations above.
• items (required): array of values. Plain values work for the common case; objects give full control. Fields set in an object item override options.
• options (optional): defaults for every item.

Output

One dataset item per input item:

{
  "index": 0,
  "operation": "bahtText",
  "input": 123.5,
  "status": "ok",
  "result": { "amount": "123.50", "baht": 123, "satang": 50, "textTh": "หนึ่งร้อยยี่สิบสามบาทห้าสิบสตางค์", "text": "หนึ่งร้อยยี่สิบสามบาทห้าสิบสตางค์" },
  "error": null
}

status is ok or invalid. When it is invalid, result is null and error explains what to fix.

---

Operations and examples

bahtText: amount in words (บาทถ้วน)

Item: a number or a string ("1,234.50", "฿1,000", Thai digits "๑๒๓"), or {"amount": ..., "language": "th" | "en" | "both"}.

Input	text
0	ศูนย์บาทถ้วน
21	ยี่สิบเอ็ดบาทถ้วน
101	หนึ่งร้อยเอ็ดบาทถ้วน
1000000	หนึ่งล้านบาทถ้วน
21000000	ยี่สิบเอ็ดล้านบาทถ้วน
0.25	ยี่สิบห้าสตางค์
"100.50"	หนึ่งร้อยบาทห้าสิบสตางค์
-5	ลบห้าบาทถ้วน
{"amount": 1234.5, "language": "en"}	One Thousand Two Hundred Thirty-Four Baht and Fifty Satang

Rules: เอ็ด for a final 1 after higher digits (11, 21, 101, 1,000,001), ยี่สิบ for 20s, ล้าน repeated for very large numbers (ล้านล้าน). Amounts are rounded half-up to satang.

thaiDate: parse and convert dates

Item: a date string, or {"date": "...", "twoDigitYear": "auto" | "be" | "ce"}.

Understands "21 มิ.ย. 67", "1 มกราคม 2569", "วันที่ 1 ต.ค. 2568", "2569-09-09" (Buddhist Era ISO), "2026-09-09", "9/9/2569" (day/month/year), Thai digits, "21 June 2024". Years ≥ 2400 are read as Buddhist Era. Two-digit years: auto reads 50–99 as BE (69 → 2569) and 00–49 as CE (26 → 2026).

Result for "21 มิ.ย. 67":

{ "iso": "2024-06-21", "yearBE": 2567, "weekdayTh": "ศุกร์",
  "thaiLong": "วันศุกร์ที่ 21 มิถุนายน พ.ศ. 2567", "thaiMedium": "21 มิถุนายน 2567",
  "thaiShort": "21 มิ.ย. 67", "thaiNumeric": "21/06/2567", "isoBE": "2567-06-21",
  "english": "Friday, 21 June 2024", "fiscalYearBE": 2567, "fiscalQuarter": 3 }

The fiscal year is the Thai government fiscal year (1 October to 30 September).

holidays: list holidays

Item: a year (2026 or 2569), "2026-04-01/2026-04-30", or {"start": ..., "end": ...}. Options: calendar = bank (default, Bank of Thailand financial-institution holidays) or government; province to include regional special days; includeWeekendHolidays (default true).

{ "operation": "holidays", "items": [2027], "options": { "calendar": "bank" } }

Each holiday has date, nameTh, nameEn, type (holiday, substitution or special), calendars, and where relevant substitutesFor, region and provisional.

businessDays: business-day maths

Item objects (action defaults to isBusinessDay; a plain date string also means isBusinessDay):

Action	Fields	Result
isBusinessDay	date	isBusinessDay, reason (holiday name or weekend), nextBusinessDay
add	date, days (negative goes back)	result, holidaysSkipped
count	start, end, includeStart/includeEnd (default true, like Excel NETWORKDAYS)	businessDays, weekdayHolidays
next / previous	date, inclusive (default true)	result

{ "operation": "businessDays",
  "items": [ { "action": "add", "date": "2026-12-30", "days": 3 },
             { "action": "count", "start": "1 เม.ย. 2569", "end": "30 เม.ย. 2569" } ] }

→ 2027-01-06 (skips New Year's Eve, New Year's Day and the weekend) and 18 business days in April 2026. Business days are Monday–Friday that are not holidays on the chosen calendar. Data covers 1 Jan 2024 – 31 Dec 2027; dates outside return invalid.

vat: VAT 7%

Item: a number (net amount, VAT added) or {"amount": 107, "mode": "inclusive"} to extract VAT from a total. rate overrides 7.

Input	amountBeforeVat	vat	totalIncludingVat
100	100.00	7.00	107.00
{"amount": 1000, "mode": "inclusive"}	934.58	65.42	1000.00
21.5	21.50	1.51	23.01

Rounding is half-up to 0.01 THB (not banker's rounding), and base + VAT always equals the total.

withholdingTax: ภาษีหัก ณ ที่จ่าย

Item: {"amount": 10700, "type": "services", "vatMode": "inclusive"} → base 10,000, VAT 700, withholding 300, net payment 10,400, form ภ.ง.ด.53.

type	Company payee	Individual payee
services (ค่าบริการ)	3%	3%
hireOfWork (ค่าจ้างทำของ)	3%	3%
professionalFees (วิชาชีพอิสระ)	3%	3%
rent (ค่าเช่า)	5%	5%
advertising (ค่าโฆษณา)	2%	2%
transport (ค่าขนส่ง)	1%	1%
nonLifeInsurance (เบี้ยประกันวินาศภัย)	1%	–
promotionPrizes (รางวัล ส่วนลด ส่งเสริมการขาย)	3%	3%
contestPrizes (รางวัลประกวด ชิงโชค)	5%	5%
publicEntertainer (นักแสดงสาธารณะ)	5%	5%
royalties (ค่าลิขสิทธิ์)	3%	–
interest (ดอกเบี้ย)	1%	15%
dividends (เงินปันผล)	10%	10%
governmentPayer (หน่วยงานรัฐจ่าย)	1%	–
foreignCompany (จ่ายนิติบุคคลต่างประเทศ)	15%	–

Other fields: payeeType (company default, or individual), vatMode (none default, exclusive, inclusive); tax is withheld on the amount before VAT. rate sets a custom rate. eWithholding: true applies the reduced 1% e-Withholding rate where it is available. Sources: Revenue Department order ท.ป. 4/2528 (as amended) and Revenue Code ss. 50, 69 tri and 70. Each result includes its legal basis. Below 1,000 THB per contract, withholding is generally not required; the result adds a note when this applies.

taxIdCheck: 13-digit ID check

Item: "0105556123456", "0-1055-56123-45-6", or {"id": "..."}. It checks the mod-11 check digit and describes the format:

• Juristic IDs (start with 0), 0-PP-T-YYY-NNNNN-C: registration province (PP), legal form (T: 2 ordinary partnership, 3 limited partnership, 5 company limited, 7 public company), registration year in Buddhist Era (YYY), running number and check digit. 0-99-… IDs are Revenue Department numbers for entities that are not registered companies.
• Personal IDs (start with 1–8): the category only. The ID is masked in the output (1XXXXXXXX3451) unless you set maskPersonalIds: false.

Nothing is looked up online and nothing is stored apart from the run's own dataset.

tokenize: Thai word segmentation

Item: text, or {"text": "...", "separator": "|", "keepWhitespace": false}. "ฉันรักภาษาไทย" → ["ฉัน", "รัก", "ภาษาไทย"]. It uses PyThaiNLP's newmm dictionary tokenizer plus all Thai subdistrict, district and province names, so addresses segment well.

romanize: Thai → RTGS

Item: text, or {"text": "...", "capitalize": true}. "หาดใหญ่" → hat yai, "กรุงเทพมหานคร" → krung thep maha nakhon, "ถนนสุขุมวิท" → thanon sukhumwit. Place names come from the official English names in the address dataset. Other words go through PyThaiNLP's rule-based royin romaniser, with fixes for silent ห and silent final ร. This is automatic RTGS: it is good for search keys, slugs and labels, but it is not a dictionary transcription (for example สวัสดี gives swatdi, not sawatdi).

address: Thai address parsing and postcode validation

Item: an address string, or a bare 5-digit postcode to look up.

{ "operation": "address",
  "items": ["99/1 ม.3 ต.บางพลีใหญ่ อ.บางพลี จ.สมุทรปราการ 10110"] }

→ houseNumber 99/1, moo 3, subdistrict บางพลีใหญ่ (Bang Phli Yai), district บางพลี (Bang Phli), province สมุทรปราการ (Samut Prakan), postcodeValid: false, suggestedPostcode: "10540".

• Fields: houseNumber, moo, village (หมู่บ้าน), building (อาคาร), floor, room, trok, soi, road, subdistrict, district, province, postcode, their English names, and unparsed (for example a company name).
• Address forms: labelled (ต./อ./จ., แขวง/เขต, full words) and unlabelled ("12 สีลม บางรัก กรุงเทพ 10500").
• Postcodes: a missing postcode is filled when it is unique. A wrong postcode is flagged with the right one.
• Ambiguity: when an address is ambiguous, candidates lists the possible areas.

---

Data sources and licences

• Holidays (src/data/holidays.json, verified 2 Nov 2026):
  • Bank calendar: the Bank of Thailand's official calendar and notifications (FPG. 8/2566, 12/2566, 4/2567, 5/2567, 31/2568, 26/2569, 37/2569). It matches BOT's published list exactly for 2024–2027.
  • Government calendar: Cabinet Secretariat circular นร 0505/ว 497 and Cabinet resolutions as reported on thaigov.go.th and prd.go.th.
  • Lunar holidays: Makha, Visakha and Asarnha Bucha use the officially announced dates. They are not computed.
  • 2027 government calendar: provisional until the Cabinet publishes it, and Royal Ploughing Day 2027 is not yet announced. The 2027 bank calendar is final.
• Addresses: kongvut/thai-province-data (MIT licence, © 2025 Kongvut Sangkla), 77 provinces, 7,436 subdistricts with postcodes. The licence is in src/data/LICENSE-address-data.txt.
• NLP: PyThaiNLP 5.4 (Apache-2.0), newmm tokenizer and royin romaniser. The image ships only the word, syllable and stopword lists.

Limits

• Holiday data: covers 2024–2027 only. Ad-hoc special holidays announced after 2 Nov 2026 are not included until the data is updated.
• Southern provinces: local bank closures for Eid and Chinese New Year are not modelled.
• Private employers: their own holiday lists are not modelled.
• Withholding tax:
  • It covers common cases for Thai companies and individuals.
  • Tax treaties, foundations and associations (10%), and progressive personal rates are not modelled.
  • Check the current e-Withholding rate for your payment date.
• Address parsing: rule-based, built for Thai-language addresses. English addresses are not parsed.
• Romanisation: automatic, so expect occasional spelling differences from official RTGS.
• Not tax or legal advice: check important filings with the Revenue Department or your accountant.

Using it from an AI agent (MCP)

Add the Actor to the Apify MCP server (https://mcp.apify.com?tools=chatuporn/thai-toolkit-for-ai-agents). The input schema describes each operation, so the model can pick the right one, for example "How many bank business days are there between 1 and 30 April 2026?" or "Write 21,500.75 baht in Thai words for this invoice."