Receipts to JSON, in any language

What we extract

Every field, with confidence and citations

Every field comes back with a confidence score in [0,1] and a normalized bounding box on the source page. Line-item totals and tax breakdown are reconciled against the printed total before the response is returned.

{
  "schema_version": "receipt.v1",
  "result": {
    "merchant_name": "Café Belga",
    "merchant_address": "Place Eugène Flagey 18, 1050 Ixelles, Belgium",
    "merchant_vat_id": "BE0456789012",
    "transaction_datetime": "2026-04-22T13:47:00+02:00",
    "currency": "EUR",
    "line_items": [
      {
        "description": "Croque-monsieur",
        "quantity": "1",
        "unit_price": { "amount": "12.50", "currency": "EUR" },
        "line_total": { "amount": "12.50", "currency": "EUR" },
        "tax_rate": "0.12",
        "category": "food",
        "confidence": 0.96,
        "bbox": { "page": 1, "x": 0.08, "y": 0.31, "w": 0.84, "h": 0.022 }
      },
      {
        "description": "Salade chèvre chaud",
        "quantity": "1",
        "unit_price": { "amount": "16.00", "currency": "EUR" },
        "line_total": { "amount": "16.00", "currency": "EUR" },
        "tax_rate": "0.12",
        "category": "food",
        "confidence": 0.95
      },
      {
        "description": "Stella Artois 25cl",
        "quantity": "2",
        "unit_price": { "amount": "4.20", "currency": "EUR" },
        "line_total": { "amount": "8.40", "currency": "EUR" },
        "tax_rate": "0.21",
        "category": "alcohol",
        "confidence": 0.97
      },
      {
        "description": "Café espresso",
        "quantity": "2",
        "unit_price": { "amount": "2.80", "currency": "EUR" },
        "line_total": { "amount": "5.60", "currency": "EUR" },
        "tax_rate": "0.12",
        "category": "beverage",
        "confidence": 0.98
      }
    ],
    "subtotal": { "amount": "37.62", "currency": "EUR" },
    "tax_breakdown": [
      { "rate": "0.12", "taxable_amount": "30.45", "amount": "3.65" },
      { "rate": "0.21", "taxable_amount": "6.94",  "amount": "1.46" }
    ],
    "tip": { "amount": "0.00", "currency": "EUR" },
    "total": { "amount": "42.50", "currency": "EUR" },
    "payment_method": "card_visa",
    "validation": {
      "totals_reconcile": {
        "valid": true,
        "computed_total": "42.50",
        "declared_total": "42.50",
        "diff": "0.00",
        "tolerance": "0.01"
      }
    }
  },
  "field_metadata": {
    "merchant_vat_id": { "confidence": 0.94, "checksum_valid": true },
    "total.amount":    { "confidence": 0.99 }
  }
}

Domain-specific validation

What makes this a specialist

Format coverage

Tested across six receipt categories, EU + US

6 categories · 40+ languages · printed and handwritten totals · thermal, laser, and PDF

Code recipes

From document to JSON in five lines

curl -X POST https://eu-api.tryparsr.dev/v1/parse/receipt \
  -H "Authorization: Bearer $PARSR_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "document_url": "https://example.com/cafe-brussels.heic",
    "wait": 30
  }'

import os, httpx

resp = httpx.post(
    "https://eu-api.tryparsr.dev/v1/parse/receipt",
    headers={
        "Authorization": f"Bearer {os.environ['PARSR_API_KEY']}",
        "Idempotency-Key": os.urandom(16).hex(),
    },
    json={
            "document_url": "https://example.com/cafe-brussels.heic",
        "wait": 30,
    },
    timeout=40,
)
result = resp.json()["result"]
reconcile = result["validation"]["totals_reconcile"]
if not reconcile["valid"]:
    raise ValueError(f"Totals don't reconcile — diff {reconcile['diff']}")
for item in result["line_items"]:
    print(item["description"], item["line_total"]["amount"])
print("TOTAL:", result["total"]["amount"], result["currency"])

const resp = await fetch("https://eu-api.tryparsr.dev/v1/parse/receipt", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.PARSR_API_KEY}`,
    "Content-Type": "application/json",
    "Idempotency-Key": crypto.randomUUID(),
  },
  body: JSON.stringify({
    document_url: "https://example.com/cafe-brussels.heic",
    wait: 30,
  }),
});
const { result } = await resp.json();
if (!result.validation.totals_reconcile.valid) {
  throw new Error(
    `totals don't reconcile — diff ${result.validation.totals_reconcile.diff}`,
  );
}
console.log(`${result.merchant_name}: ${result.total.amount} ${result.currency}`);

from langchain_parsr import ParsrToolkit
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

tools = ParsrToolkit.from_env().get_tools()
agent = create_react_agent(ChatOpenAI(model="gpt-4o"), tools)

result = await agent.ainvoke({
    "messages": [(
        "user",
        "Parse this lunch receipt and categorize it for our expense system: "
        "https://example.com/cafe-brussels.heic"
    )]
})
print(result["messages"][-1].content)

What teams build with this

Three integration patterns

Compared

How parsr's receipt parsing compares

FAQ

Common questions

• Will phone photos work, or do I need a flatbed scan?

  Phone photos work. The vision LLM tolerates skew, glare, and creases down to roughly 150 DPI equivalent (a typical iPhone photo of an A6 receipt held in hand). Confidence drops on heavy motion blur, deep shadows across the totals line, or paper folded across columns — surfaced through field_metadata.<field>.confidence so your agent escalates only when needed. HEIC, JPEG, PNG, PDF, and multi-page TIFF all in.

• How well does it handle non-English / non-Latin receipts?

  40+ languages out of the box, including Polish, Czech, Hungarian, Greek, and Turkish alongside the usual EN/FR/DE/ES/IT/NL/PT. Field names and tax-rate plausibility are jurisdiction-aware — a Greek receipt is checked against Greek VAT rates (24% / 13% / 6%), a Czech receipt against Czech rates (21% / 12%). Cyrillic and CJK receipts are supported but accuracy on those scripts is in the 88–92% range rather than 95%+; we surface that honestly via per-field confidence.

• What if a receipt has multiple currencies on it?

  Duty-free and border-shop receipts often print prices in two currencies (e.g. EUR + CHF, or EUR + GBP). The parser returns receipt-level currency as the dominant one and per-line currency on each line_item where it differs. validation.currency_consistency reports drift instead of silently coercing — your integration can decide whether to sum, convert at the printed rate, or escalate.

• What about faded thermal-paper receipts?

  Thermal paper that's been in a wallet for a few weeks is the hardest case. parsr extracts what's legible and lowers confidence on the rest — typical accuracy on aged thermal sits in the 80–90% range vs 95%+ on fresh prints. Use field_metadata.<field>.confidence < 0.85 as the recommended escalation threshold; do not treat low-confidence fields as ground truth. We don't claim accuracy we can't deliver on bad inputs.

• Some of my receipts are just a total — no line items. Does that still work?

  Yes. line_items returns an empty array, and merchant, total, currency, payment_method, and tax_breakdown are populated from whatever the receipt prints. Totals reconciliation simply checks tax_breakdown + tip = total when there's no subtotal to verify against. Common case: parking tickets, vending receipts, and bar tabs that only print the round amount.

• Does it support iPhone HEIC straight from the camera roll?

  Yes. HEIC, HEIF, JPEG, PNG, WebP, PDF, TIFF — uploaded as multipart/form-data or as a public URL. iOS shares HEIC by default and we don't ask users to convert before upload. EXIF rotation is honored. Live Photos are accepted as the still frame. Max 25MB per file on the standard tier; larger via signed URLs.