Extract invoice data from PDFs without retyping it
The PDF Invoice Extractor reads an invoice and fills in the fields you need for bookkeeping: invoice number, invoice date, due date, vendor, customer, the seller’s tax ID (GSTIN, VAT number or EIN), currency, subtotal, tax and total. It works with digital PDFs from billing software, scanned invoices and phone photos, and it checks the arithmetic for you before anything reaches your books.
The process is deliberately transparent. First the text of the invoice is extracted in your browser — from the PDF’s text layer, or with OCR for scans and photos — and shown so you can see and correct it. Only if you then press Read fields with AI is that plain text sent for field extraction. The AI cannot make values up: code on Galaxy of Tools’ server lists every amount, date, ID and name line that actually appears in the text, and TypeSafe’s Jev model only chooses which of those candidates plays each role, with a confidence score.
Every field stays editable, low-confidence values are marked please check, and a running table exports to CSV for Excel or your accounting software.
How to extract data from an invoice
- Drop an invoice PDF, a photo or a screenshot onto the upload area. For photos and scans, choose the OCR language first if the invoice is not in English.
- Check the Text of the invoice box. Fix any obvious OCR mistakes in amounts — the fields can only be as good as the text.
- Press Read fields with AI. Or, to keep everything offline, press Fill from keywords, which looks for labels like “Total”, “GST” and “Invoice No”.
- Review the fields. Green percentages are confident picks; amber please check badges need a look. The check line below the fields confirms whether subtotal + tax equals the total.
- Press Add to table and repeat for your other invoices, then Download table as CSV. You can also save the current invoice as JSON or CSV.
PDF Invoice Extractor features
Works with PDFs, scans and photos
Digital PDFs are read from their text layer; scanned PDFs, JPG, PNG and WebP images are read with Tesseract OCR in your browser.
AI that can only pick, not invent
Candidates come from the document text itself, so every value the AI returns is a span that really appears on the invoice.
Confidence on every field
Each value shows the model’s confidence; anything under 60% is flagged “please check” and outlined in amber.
Built-in arithmetic check
Subtotal + tax is compared with the total in code, with a tolerance for rounding, and a due date earlier than the invoice date is flagged.
Bookkeeping-ready CSV
Amounts become plain numbers (₹1,180.00 → 1180), and dates get an extra ISO column (2026-03-05) that sorts correctly in any spreadsheet.
Private by default
Nothing is sent unless you ask, the file itself never leaves your device, and the running table is stored only if you tick “Remember”.
Which invoice fields are extracted
| Field | What it looks for | Typical labels |
|---|---|---|
| Invoice number | The document’s own reference — not a PO, order or phone number | Invoice No, Inv #, Bill No |
| Invoice date / due date | Issue date and payment due date | Date, Invoice Date, Due Date, Payment Due |
| Vendor / customer | Seller name and the billed party | From, Bill To, Sold To, Buyer |
| Seller tax ID | GSTIN (India), VAT number (UK/EU), EIN or TIN (US) | GSTIN, VAT Reg No, Tax ID |
| Subtotal, tax, total | Amount before tax, total tax and the final amount payable | Subtotal, Taxable Value, CGST/SGST/IGST, VAT, Grand Total, Amount Due |
| Currency | From symbols, codes, tax names and addresses | ₹, Rs, $, £, €, INR, USD, GBP |
Getting reliable results
- Prefer the original PDF over a photo — its text is exact, so OCR errors cannot creep in.
- Split GST: when an Indian invoice shows CGST and SGST separately, check that the tax field holds the combined amount; if only the halves are printed, add them and type the total.
- Totals that don’t add up are often legitimate — discounts, shipping, TDS, round-off or a second tax line. The check simply tells you where to look.
- Dates such as 05/03/2026 are read day-first (5 March) unless the currency is US dollars, when month-first is used. The original text is always kept next to the ISO date.
- For line items rather than header fields, use the PDF Table Extractor; for till receipts, the PDF Receipt Extractor; to create invoices, the Invoice Generator.
What is sent, and what is not
Text extraction, OCR, the keyword fill, the arithmetic check, editing and exports all happen in your browser. Pressing “Read fields with AI” sends one request with the invoice text (up to 12,000 characters) to Galaxy of Tools’ server, which asks TypeSafe’s Jev model and returns the chosen fields. The file is never sent, and neither the text nor the results are stored. The extracted figures are an aid for bookkeeping, not tax or accounting advice — always compare them with the original invoice before you file a return or make a payment.
PDF Invoice Extractor FAQ
How do I extract data from an invoice PDF?
Upload the invoice, check the extracted text, press “Read fields with AI” (or “Fill from keywords” to stay offline), review the fields and export them as CSV or JSON. The PDF invoice extractor is free and needs no account.
Is my invoice uploaded or stored?
The file never leaves your device. Only the extracted text is sent, and only when you press “Read fields with AI”; it is used for that one request and not stored.
Can it read scanned or photographed invoices?
Yes. Scanned PDFs and images are read with OCR in your browser. Sharp, straight, well-lit images give the best results; always check amounts after OCR.
Does it support GST and VAT invoices?
Yes. It finds GSTIN and VAT numbers, GST, CGST/SGST/IGST and VAT amounts, and currencies such as ₹, $, £ and €.
What does “please check” mean?
The model’s confidence for that field was below 60%, usually because several similar values appear (for example two dates). Compare it with the invoice and correct it if needed.
Can I process many invoices at once?
Add invoices one after another to the table; it keeps every row until you clear it, and “Download table as CSV” exports them all. Tick “Remember” to keep the table in this browser between visits.
What if the AI is unavailable?
You will see a message, and everything else keeps working: the text stays in the box, the keyword fill works offline, and you can type or correct the fields yourself.