Local Invoice Processor

Problem

When reimbursement files are scattered across folders, manually checking each invoice, verifying amounts and tax fields, then transcribing them into Excel is repetitive and error-prone. Invoices may arrive as PDF, JPG, PNG, or WebP, so the workflow needs OCR, field extraction, and category-based organization. This is common in finance pre-review, reimbursement preparation, and archiving workflows.

How It Works

The skill processes a folder of invoice files in batches. It skips non-invoice documents, runs OCR on image or PDF content, and extracts fields such as invoice number, date, issuer, amount, and tax amount. It then classifies invoices using keywords into categories such as catering, office equipment, transport, lodging, or communications, and generates an Excel summary with formulas plus categorized folders in the source directory. The skill description also mentions optional local LLM assistance for field extraction, while the core workflow remains local.

Boundaries

It works best with clear, standard-format invoices; the skill description notes 95%+ recognition for standard layouts, while handwritten, blurry, or irregular invoices may need manual review. Processing is local and does not upload data. First runs may be slower while the OCR model loads; Windows and Linux require PaddleOCR, while macOS can use the system Vision engine.