One scan in. Every document out, split and named.
Drop the whole stack from the scanner. Bindery proposes where each document starts, you check the cuts, and every document comes out as its own named PDF.
Free: finding the cuts uses no job. Downloading the documents counts as 1 of 3 daily jobs, files go up to 25 MB, and PDF exports include a visible Made with Bindery watermark. Pro removes the size cap and the watermark.
Prefer the full studio? Open Unstack in the app →
A sheet-fed scanner or a phone-scan session turns a pile of paper into one long PDF: three invoices, a bank statement, a letter and the blank backs between them. Unstack reads that PDF page by page in your browser and proposes a cut wherever the evidence says a new document starts: a page that says “Page 1 of 3”, the page after one that said “3 of 3”, a blank separator sheet, a new invoice or reference number labelled at the top of the page, a new letterhead, a new labelled date, a header and footer unlike the page before, or a change of orientation or page size. Evidence that pages belong together counts too: “Page 2 of 3” joins the page before and so does the same labelled number; a page that begins “continued” joins the page before; one that ends “continued on next page” joins the page after. Pages with text are read as they are; on a scanned page only a strip at the top and the bottom is read, with OCR on your device that does not use your free OCR pages.
Every proposed cut is drawn between the page thumbnails — solid when the evidence is strong, dashed when it is a guess — and its tooltip lists the reasons, such as a page 1 of N on the next page or a new sender. One click adds or removes a cut, and blank pages are greyed out and left out unless you click to keep one. Each document gets a type — invoice, receipt, bank statement or letter, and with Tax season on, W-2s, 1099s and other tax forms — and a name from its own date, type and sender, such as 2025-01-31_Invoice_Acme_Supplies_Ltd, which you can type over. Download a ZIP with one PDF per document and an index CSV, or one PDF with a bookmark per document. Pages are copied, not re-rendered, and nothing is uploaded.
How it works
- Drop the scanned PDF and open Unstack — the cuts are worked out page by page on your device, with progress and a Cancel button
- Check the cuts between the thumbnails: click to add or remove one, click a greyed-out blank page to keep it, and fix any name or type
- Download a ZIP of named PDFs with unstack-index.csv, or one PDF with a bookmark per document
What decides a cut
| Evidence | What it does | How it is read |
|---|---|---|
| Blank page | Cuts after the blank page, or once after a run of them, and leaves the blanks out of the documents — unless the next page says page 2 of N or later, or begins “continued” | The share of dark pixels on a small render of the page, ignoring a thin border where scanners leave edges |
| Page 1 of N, or N of N | A strong cut: before a page 1, after a last page | “Page 1 of 3”, “Seite 1 von 3”, “Página 1 de 3”, “Page 1 sur 3”, “Pagina 1 di 3”, or a bare “1/3” alone on a line at the top or bottom of the page |
| Page 2 of N or later | A strong join with the page before, even when a number on the page has changed | The same forms |
| “Continued” | A strong join | A page that begins “continued”, “Fortsetzung”, “continuación” or “(suite)” joins the page before; one whose foot says “continued on next page”, “Fortsetzung folgt”, “suite au verso” or “continúa en la página siguiente” joins the page after |
| Document number | A different number cuts; the same number joins | Only a number after a label such as Invoice No, Receipt #, Order number or Reference in the top 30% of the page, so transaction references further down a statement are ignored |
| Sender | A new one counts toward a cut | The letterhead: the largest line near the top, and only when it is clearly larger than the body text |
| Date | A new one counts toward a cut | Only a labelled date, such as Invoice date or Statement date — not the first date on the page |
| Document type | A new one counts toward a cut | Phrases near the top of the page that mark an invoice, a receipt, a bank statement or a letter; with Tax season on, a tax form title in the top or bottom strip |
| Header and footer | A different one counts toward a cut | The lines in the top 12% and bottom 10% of the page compared with the page before, with the text size and left margin |
| Orientation and size | A change counts toward a cut | The page as it is displayed, after any rotation |
Strong evidence gives a solid cut. The weaker kinds add up, and a cut that rests on them is drawn dashed: a guess, worth a look. Each cut’s tooltip lists its reasons. One click adds or removes a cut; from the keyboard, ← and → move between the gaps and Space toggles one.
Names, types and the index
Each document is named from its own pages with the pattern {date}_{type}_{from} — the document date, its type and its sender — so an invoice comes out as 2025-01-31_Invoice_Acme_Supplies_Ltd. A field that is not found drops out with the underscore before it, accents are folded, anything other than a letter, digit or hyphen becomes an underscore, and when two documents would get the same name the second becomes name_2. A Social Security, employer identification, card or bank account number, or an email address, is taken out of a name before it is used, whether Bindery proposed the name or you typed it. The type comes from the document’s first page; change it from the list next to the name if it is wrong.
The ZIP carries unstack-index.csv, one row per document: name, pages, type, date, sender, number, total and confidence. The same numbers are kept out of it, and a value a spreadsheet would run as a formula is written as plain text. Keep the index with the documents: next time, Compare with last year’s index reads it back and lists the types and senders that were in last year’s stack and are missing from this one, matched by type, and by sender where one was read.
What comes out
- Pages copied, not redrawn. Each document’s pages are copied from the stack, so a scan keeps its resolution unless you choose compression, and a page with text keeps its selectable text; on the free plan each page also carries the small “Made with Bindery” line.
- A ZIP or one PDF. One PDF per document with unstack-index.csv, or the whole stack as one PDF with a bookmark per document and the blank pages left out.
- Searchable if you want. A full text layer can be added to scanned documents with OCR on your device; those pages count against the 5 free OCR pages a day, which Pro removes. Reading the strips to find the cuts does not.
- Next steps. From the done screen, hand the documents to Application pack, Bank statement → Excel or Rename by content.
Honest limits
- It proposes; you decide. Guesses are shown dashed. Look at those before you download, and at any name that looks wrong, which often means a missed cut.
- Separator sheets help. A blank sheet between documents is the clearest evidence there is. A blank page counts as a cut unless the page after it says page 2 of N or later or begins “continued”, so a one-sided letter scanned in duplex may still be cut at each blank back.
- Show-through. A blank back with heavy show-through from the other side can count as a page with content.
- Scans are read in strips. On a page without text only about the top fifth and the bottom tenth are read with OCR, so a clue halfway down the page is not seen, and a letterhead that is only a logo gives no sender.
- Look-alike documents. Two letters from the same sender with no page numbers, labelled numbers or dates can read as one document; cut them by hand.
Splitting in the browser versus uploading
| Unstack | Bindery | Upload-based tools |
|---|---|---|
| Where the file is processed | In your browser tab, on your device | Uploaded to the service’s servers |
| The stack | Read and split on your device | Uploaded in full, with every document in it |
| Checking the cuts | Every cut shown between thumbnails with its reasons, changed with a click | Varies |
| Scanned pages | Top and bottom strips read with OCR on your device | OCR on the service’s servers |
| What exists afterwards | The original and the new file, both on your device | A server copy until the retention window ends |
| Works offline | Yes, once the app has loaded | No |
Questions people ask
How does Bindery decide where one document ends?
It weighs the evidence on every page and shows its working. Page numbers, blank separator sheets, invoice and reference numbers labelled at the top of a page, letterheads, labelled dates, the lines in the header and footer, and each page’s orientation and size count for or against a cut; a solid cut rests on strong evidence and a dashed one is a guess for you to check. Page numbers are recognised in English, German, Spanish, French, Italian and Portuguese, such as “Seite 2 von 5”, “Página 2 de 5” and “Page 2 sur 5”, and “continued” in English, German, French and Spanish. Nothing is written until you download, so a wrong cut costs one click.
What goes into the file names?
The pattern {date}_{type}_{from}: the document’s own date, its type and its sender, as in 2025-01-31_Invoice_Acme_Supplies_Ltd. A field that is not found drops out with the underscore before it, two documents never get the same name, and every name can be typed over before you download. A Social Security, employer identification, card or bank account number, or an email address, is taken out of a name before it is used, so a sender line or a number that carries one never ends up in a file name.
Is my stack uploaded?
No. The pages are read, the cuts worked out and the documents written in your browser tab; nothing is sent anywhere, which matters for a pile of bank statements, tax forms and letters. Finding the cuts uses no job; downloading the documents counts as one of your 3 free jobs a day, however many documents the stack holds. Free files can be up to 25 MB, and free PDFs carry a small “Made with Bindery” line on each page, including the PDFs in the ZIP; Pro removes both.
Does it work on scans with no text?
Yes. A page with fewer than 40 characters of text, such as a plain scan or a scan whose only text is a scanner stamp or a Bates number, has its top and bottom strips read with OCR on your device. That is enough for page numbers, letterheads, labelled numbers and dates. It does not use your 5 free OCR pages a day; only the optional searchable text layer on the documents you download does.
What does Tax season do?
It adds US tax forms to the types Bindery recognises: W-2, 1099-INT, 1099-DIV, 1099-NEC, 1099-MISC, 1099-B, 1099-R, 1098, 1098-T, 1098-E, 1095-A, Schedule K-1 and SSA-1099. A form counts only by a form title in the top or bottom strip of the document’s first page, so a 1040 that says “Attach Form(s) W-2 here” is not taken for a W-2. With it off, those forms are typed as Document; invoices, receipts, bank statements and letters are recognised either way.
Can I keep a blank page?
Yes. Blank pages are greyed out and left out of the documents; click one to keep it. A run of blank pages belongs, left out, to the document before it, and there is one cut after the run, so one click joins the documents on either side of it.