1. Documentor
  2. Document sorting
Intake · Archive suggestion · Triage · QR codes and barcodes · Labels

A document arrives and finds its own archive.

The optional Intake plugin takes documents from a shared folder, another system or e-mail and suggests which archive they belong to. A sorter confirms the suggestion or, after a trial period, leaves it to automation. Inside the archive, labels then sort the documents – from a language model as well as by QR code and barcode.

How a document travels

From an e-mail attachment to search in the archive

  1. 01

    Receipt

    The file arrives in a sorting inbox from a folder in S3, through the API or as an e-mail attachment. Its format and size are checked; the same file is recognised by its content and not accepted twice.

  2. 02

    Preparation

    A preview of the first page, the text of the first pages (through OCR for a scan) and QR codes and barcodes from the first three pages. A format other than PDF is converted.

  3. 03

    Archive suggestion

    From the sender, by a code rule, by similarity to the documents in the target archives, or from a language model – in that order.

  4. 04

    Triage

    The sorter confirms the suggestion, picks another archive or rejects the document, also in bulk. With automation switched on, a certain document is filed on its own.

  5. 05

    Filing

    The document is created in the archive and processed like any other. The sender and subject of the e-mail stay with it, marked unverified.

Rules by code

A QR payment code spots an invoice before a person does.

An invoice with QR Platba (the Czech QR payment code), a job ticket with the job’s barcode, a separator sheet from the scanning line – the code on the first page says where a document belongs more precisely than its text. An inbox has “pattern → archive” rules, and the first one from the top that matches decides.

  • Patterna regular expression anywhere in the code text; matching is case-sensitive, (?i) turns that off
  • Code typeoptionally QR, Data Matrix, EAN, Code 128 and others – the rule then ignores other codes
  • Archiveonly from the inbox’s target archives; an archive that a rule points to cannot be removed from the inbox
  • What to go bythe detail of an incoming document shows its codes – you write the pattern from the real values
DocumentPatternArchive
invoice with QR Platba^SPD\*Received invoices
payment to a subsidiary’s accountACC:CZ65…Invoices – Service
job ticket^ZAK-\d{5}$Jobs
batch separator sheet^ARCHIV/SMLOUVY/Contracts
Triage

A person decides until you trust the suggestions.

The Incoming page shows what is waiting for a decision, with the suggestion and a confidence label. The number in the menu says how much there is, and the list refreshes itself.

  • One clickConfirm, Elsewhere… or reject; ticked rows all at once
  • Detailpreview, text excerpt, codes from the first pages, sender and subject; open the PDF or download the original
  • Mistakesa rejected document can be returned to triage for about 30 days
  • Who sortsthe company-wide sorter; an archive’s Processor only into archives where they may process; the Supervisor manages the inboxes
Incoming documents awaiting triage: nine documents from S3, the API and e-mail, five with a suggested archive labelled code rule, two with an uncertain suggestion by similarity and two without a suggestion, each with the buttons Confirm or File, Elsewhere and Reject
Incoming · archive suggestion and triageIntake
Automatic filing

Automation only once the suggestions have earned it.

Automation is a property of the inbox and is switched off by default. A document filed automatically shows Decided by: automation in its detail.

Switching it on

From a chosen confidence

A document whose suggestion reaches the chosen confidence (70–100 %, 90 % by default) is filed without a sorter. An archive chosen by the sender, a code rule and a single target are always filed.

Safeguards

What is never filed automatically

An e-mail attachment – both the sender and the text can be forged, and anyone can write. A document that could not be read in full. A document whose archive was chosen by the language model.

Before you switch it on

Agreement of suggestions first

Let the inbox run with manual confirmation and watch how often sorters confirmed the suggested archive, overall and by confidence label. Only then switch on automation.

Inputs

From a folder, another system or e-mail.

An inbox is a mailbox for documents of one kind, such as Received invoices. It defines which archives its documents can be filed into and where they come from. The formats are the same as for uploads: PDF, Word, XPS, Markdown and images.

S3

A folder for the scanner and batches

A scanning line or a tool such as rclone uploads files into the inbox folder. Suitable for large files and batches; PDFs up to 2 GB.

API

Another system with a key

One request per file; the inbox key can only upload and check the status of its own files. Optionally with the code of the target archive; resending after an outage is safe.

E-mail

Every attachment is a document

An IMAP or Microsoft 365 mailbox. A logo from a signature is skipped, an unsupported attachment is rejected with a reason, and the message is moved to the processed folder; nothing is deleted.

  • S3
  • REST API
  • IMAP
  • Microsoft 365
Inside the archive

Labels: invoice, order, to pay.

An archive does not have to be one big pile. Labels divide documents by kind or status, and the document list has a filter by label with counts. The detail shows where each label came from.

Language model

By content

The Archive administrator writes a list of labels, each with a description of how to recognise the document, and sets a confidence threshold and the number of labels per document. The model chooses only from the list; one call per document, an external model only with the archive’s consent.

  • invoice – a tax document with a number, a supplier and an amount to pay
  • order – a request for a delivery with an order number
  • contract – a signed agreement between two parties
QR codes and barcodes

By code, without AI

Rules of the form “pattern · code type · label”. Test shows how many documents would get the label, along with sample codes; after saving, the labels are recalculated across the whole archive without reading the pages again.

  • ^SPD\* · QR code → to pay
  • ^OBJ- · Code 128 → order
  • ^\d{13}$ · EAN-13 → goods

We will show you how it works on your own documents.

Send us a few PDFs from your archive – old scans are fine. In 30 minutes you will see how searching them works and how the chat answers.

Already a customer? Sign in