ReceiptScanner by Packr

ReceiptScanner is a small Laravel package for extracting structured receipt data from images and PDFs using upstream AI providers.

It is a wrapper around multimodal LLM APIs. It does not host its own OCR service, database, queue, UI, or REST API.

Features

• Laravel facade entry point: Jake142\ReceiptScanner\Facades\ReceiptScanner
• Public methods:
  • ReceiptScanner::scanImages(array $images): array
  • ReceiptScanner::scanPdf(mixed $pdf): array
• Supports multi-image receipt analysis in a single upstream request
• Supports PDF receipt analysis from one input file
• Provider selection via config/env
• Default provider models:
  • OpenAI: gpt-5.4-nano
  • Azure OpenAI: gpt-5.4-nano
  • Gemini: gemini-2.5-pro
  • Anthropic: claude-sonnet-4-20250514
• Configurable output fields to reduce prompt size and response size
• Safe logging controls

Requirements

• PHP ^8.3
• Laravel illuminate/support ^11.0|^12.0|^13.0
• Laravel illuminate/http ^11.0|^12.0|^13.0
• For development and testing:
  • orchestra/testbench ^9.0|^10.0|^11.0

Installation

composer require jake142/receiptscanner

Publish the config file:

php artisan vendor:publish --tag=receiptscanner-config

Configuration

ReceiptScanner is configured through config/receipt-scanner.php and environment variables.

Environment variables

RECEIPT_SCANNER_PROVIDER=openai
RECEIPT_SCANNER_TIMEOUT=60
RECEIPT_SCANNER_MAX_RETRIES=2
RECEIPT_SCANNER_PROMPT_LANGUAGE=en

OPENAI_API_KEY=
OPENAI_MODEL=gpt-5.4-nano

AZURE_OPENAI_API_KEY=
AZURE_OPENAI_ENDPOINT=
AZURE_OPENAI_DEPLOYMENT=gpt-5.4-nano
AZURE_OPENAI_MODEL=gpt-5.4-nano

GEMINI_API_KEY=
GEMINI_MODEL=gemini-2.5-pro

ANTHROPIC_API_KEY=
ANTHROPIC_MODEL=claude-sonnet-4-20250514

Config shape

The package config exposes these keys:

• default_provider
• providers.openai.api_key
• providers.openai.model
• providers.azure_openai.api_key
• providers.azure_openai.endpoint
• providers.azure_openai.deployment
• providers.azure_openai.model
• providers.gemini.api_key
• providers.gemini.model
• providers.anthropic.api_key
• providers.anthropic.model
• timeout
• max_retries
• enabled_fields
• prompt_language

Default output fields

By default, all fields are enabled:

• merchant
• total_amount
• currency
• date
• vat_amount
• mcc
• vats
• line_items
• confidence
• tip
• purchase_country
• purchase_city

You can disable parts of the response in config to reduce prompt size and response size. For example, if you do not need VAT breakdowns, set enabled_fields.vats to false.

Usage

Scan multiple images

Use scanImages() when a receipt is split across several photos.

use Jake142\ReceiptScanner\Facades\ReceiptScanner;

$result = ReceiptScanner::scanImages([
    storage_path('app/receipts/receipt-part-1.jpg'),
    storage_path('app/receipts/receipt-part-2.jpg'),
]);

The images are analyzed together as one receipt. This is useful when the top and bottom of a long receipt were captured in separate photos.

Scan a PDF

Use scanPdf() for a single PDF input.

use Jake142\ReceiptScanner\Facades\ReceiptScanner;

$result = ReceiptScanner::scanPdf(storage_path('app/receipts/receipt.pdf'));

Result format

$result is a JSON-compatible associative array. The package asks the upstream model to return strict JSON only.

Default output shape:

[
    'merchant' => null,
    'total_amount' => null,
    'currency' => null,
    'date' => null,
    'vat_amount' => null,
    'mcc' => null,
    'vats' => [
        [
            'rate' => null,
            'amount' => null,
            'amount_inc_vat' => null,
            'amount_ex_vat' => null,
        ],
    ],
    'line_items' => [
        [
            'description' => null,
            'quantity' => null,
            'unit_price' => null,
            'amount' => null,
        ],
    ],
    'confidence' => null,
    'tip' => null,
    'purchase_country' => null,
    'purchase_city' => null,
]

Notes:

• vats is always an array.
• line_items is always an array.
• Unknown scalar values are returned as null.
• Unknown arrays are returned as [].
• Dates are normalized to YYYY-MM-DD when possible.
• Numeric values are normalized to numbers, not strings, when possible.
• mcc is AI-estimated because receipts usually do not contain MCC directly.
• tip is a numeric tip/gratuity amount when visible on the receipt; otherwise null.
• purchase_country is the purchase country when inferable from receipt text, merchant/address, currency, or visible location; otherwise null.
• purchase_city is the purchase city when visible or clearly inferable from receipt text/address; otherwise null.

Provider selection

Set the provider and model in env, then load them through config.

Example:

RECEIPT_SCANNER_PROVIDER=openai
OPENAI_MODEL=gpt-5.4-nano

Azure OpenAI is supported as well. When using Azure OpenAI, configure the Azure provider settings and use the OpenAI gpt-5.4-nano model/deployment.

Logging

If logging is enabled, the package may log safe diagnostics such as provider, model, mime type, retry count, duration, and failure category.

It will not log API keys, raw receipt contents, extracted JSON values, or full base64 payloads.

Error handling

The package throws Jake142\ReceiptScanner\Exceptions\ReceiptScannerException for configuration, input, upstream, and parsing failures.

Testing

No tests are included in this package.

Disclaimer

ReceiptScanner uses upstream AI providers to interpret receipts. It does not guarantee accounting accuracy. Always review critical financial data before using it in production workflows.