EthanYoQ/Invoice-Downloader

What it solves

InvoiceFlowAI automates the tedious process of collecting electronic invoices from emails. It solves the problem of manually searching for PDF, OFD, and XML invoices across email threads, downloading them, and organizing them into folders and spreadsheets for reimbursement purposes.

How it works

The tool connects to QQ or 163 email accounts via IMAP to scan for invoices. It uses a "four-layer intelligent funnel" (whitelist domains, subject keywords, body detection, and QR code scanning) to filter out non-invoice emails.

Once an invoice is identified, the system handles it based on the type:

• Direct Attachments: Extracts PDF, OFD, or XML files.
• Links: Uses Playwright to automate the downloading of invoices from platforms like Baiwang Cloud or tax platforms.

For data extraction, it employs a three-tier AI architecture:

1. Track A (OCR + LLM): High-precision extraction of text structures.
2. Track B (GLM-4.5V Vision): Direct visual analysis for complex layouts or images.
3. Local Fallback: Regular expression rules for offline use.

Finally, it renames files based on natural language rules and generates a summary Excel report.

Who it’s for

Individuals, freelancers, and small teams who handle multiple electronic invoices monthly and need a streamlined way to prepare for reimbursement.

Highlights

• Dual-Engine AI Recognition: Automatically switches between OCR/LLM and visual recognition to maximize success rates.
• Automated Link Recovery: Automatically opens and downloads invoices from third-party tax platforms via browser automation.
• Intelligent Matching: Automatically pairs related documents, such as hotel invoices with their corresponding water bills or taxi receipts with trip itineraries.
• Natural Language Classification: Allows users to define custom sorting rules using plain language (e.g., "Put Didi rides over 100 yuan into the Large Amount folder").
• Local Privacy: Processes emails and files locally, encrypting credentials using DPAPI (Windows) or Keychain (macOS).