ticruz38/skills/receipts-ocr

Receipts OCR Skill

Extract data from receipt photos using Google Vision API. Automatically identifies merchant names, amounts, dates, and line items with confidence scoring.

Features

• OCR Processing: Uses Google Vision API for accurate text extraction
• Smart Parsing: Automatically extracts merchant, total, tax, date, and line items
• Confidence Scoring: Each field includes confidence level (0-100%)
• Category Detection: Auto-categorizes receipts (Groceries, Dining, Gas, etc.)
• Manual Correction: Update any extracted field manually
• Image Storage: Optionally copies images to local storage
• Export: Export to CSV for accounting software

Installation

cd skills/receipts-ocr
npm install
npm run build

Configuration

Uses the same Google OAuth profile as other Google services. Ensure you have connected Google OAuth first:

npm run cli -- status

Usage

CLI

# Check status
npm run cli -- status

# Scan a receipt
npm run cli -- scan receipt.jpg --copy --notes "Business lunch"

# Extract text without saving
npm run cli -- text receipt.jpg

# List all receipts
npm run cli -- list
npm run cli -- list --status pending --limit 10

# Get receipt details
npm run cli -- get 1

# Update receipt
npm run cli -- update 1 --merchant "Starbucks" --total 12.50 --category "Dining"

# Confirm/reject receipt
npm run cli -- confirm 1
npm run cli -- reject 1

# Delete receipt
npm run cli -- delete 1

# Statistics
npm run cli -- stats

# Export to CSV
npm run cli -- export receipts.csv

Library

import { ReceiptsOCRSkill } from '@openclaw/receipts-ocr';

const skill = new ReceiptsOCRSkill({ profile: 'default' });

// Process a receipt
const receipt = await skill.processReceipt('/path/to/receipt.jpg', {
  copyImage: true,
  notes: 'Business expense'
});

console.log(receipt.merchant);  // "Starbucks"
console.log(receipt.totalAmount);  // 12.50
console.log(receipt.totalConfidence);  // 0.85

// List receipts
const receipts = await skill.listReceipts({ status: 'pending' });

// Update with corrections
await skill.updateReceipt(receipt.id!, {
  merchant: 'Corrected Name',
  totalAmount: 15.00,
  status: 'corrected'
});

// Export to CSV
const csv = await skill.exportToCSV();
fs.writeFileSync('receipts.csv', csv);

await skill.close();

Data Storage

Receipts are stored in ~/.openclaw/skills/receipts-ocr/:

• receipts.db - SQLite database with receipt data and line items
• images/ - Copied receipt images (if --copy flag used)

Confidence Scoring

Each extracted field has a confidence score:

• High (80-100%): Green - Likely accurate
• Medium (50-79%): Yellow - Review recommended
• Low (<50%): Red - Manual verification needed

Supported Image Formats

• JPEG/JPG
• PNG
• GIF
• BMP
• TIFF
• WebP

Categories

Auto-detected categories include:

• Groceries
• Dining
• Gas
• Pharmacy
• Office
• Electronics
• Transportation
• Entertainment
• Retail

Schema

Receipts Table

• id - Primary key
• image_path - Path to receipt image
• merchant - Store/restaurant name
• merchant_confidence - OCR confidence (0-1)
• total_amount - Total amount
• total_confidence - OCR confidence (0-1)
• tax_amount - Tax amount
• tax_confidence - OCR confidence (0-1)
• date - Transaction date (ISO format)
• date_confidence - OCR confidence (0-1)
• category - Expense category
• payment_method - Payment type
• raw_text - Full OCR text
• status - pending/confirmed/corrected/rejected
• notes - User notes

Line Items Table

• id - Primary key
• receipt_id - Foreign key to receipt
• description - Item description
• quantity - Quantity purchased
• unit_price - Price per unit
• total - Line total
• confidence - OCR confidence (0-1)