Documents
Reading and checking documents automatically
A document goes in, and the values come out checked and with their source. Whatever the system cannot read with certainty, it puts in front of a person.
In short
Document recognition in Beyonetix OS (BEYOS) reads invoices, delivery notes, contracts and other documents from PDF, scan, photo and mail. It is meant for accounting, purchasing and administration teams that no longer want to key in large intakes by hand.
Updated: November 2026
Intake routes
How documents come in
You drag files into the window, pick them with a button or drop them into a watched folder. Mailboxes polled at fixed intervals, capture with a phone and an interface for other systems are available as well. Mailboxes, phone capture and the interface are switched on per tenant.
A mail is split into its documents. Every attachment that is a document becomes a record of its own. Logos and signature images are left out. If the mail has no attachment, the mail text itself becomes the record. Sender, subject and date stay attached to the record.
The system also unpacks ZIP archives when they are nested inside one another. It separates several documents in one scan file at reliable page breaks. If a break is unclear, it suggests the split and waits for your confirmation.
- Drag, select, folder
- Mailbox, phone, interface
- Mails and ZIP archives are split up
Formats
Which files the system understands
| Kind | Files | How it is handled |
|---|---|---|
| Including damaged PDFs | Read, repaired first when needed | |
| Images | Scan and photo, multi-page TIFF, HEIC | Each image becomes a page, large images are scaled down |
| Office formats | Word, Excel, OpenDocument, RTF | Converted to PDF, the original stays stored |
| Text | Text, CSV, HTML | Becomes a readable record |
| E-invoice | XRechnung, ZUGFeRD, also inside a PDF | The XML is authoritative and read directly |
| EML and Outlook messages | Split into documents | |
| ZIP | Archives, including nested ones | Unpacked, each file becomes a record |
Recognition in stages
How the document is read
If the file contains an XML part, as an e-invoice does, that XML is authoritative and read directly. Everything else goes through stages. First the text layer is built, for scans from text recognition. Then the system determines the document type.
Next an engine reads the document for the first time, the first read. A second, different engine reads it again independently, the second read. If both agree, the value stands. If they differ, an arbitration step decides.
First and second read run on Documents' infrastructure. An external arbitration step is optional and can be switched off per partner. Only platform administration can enable it, and only per tenant.
Flow
The stages at a glance
Checks
What is checked before handover
The system recalculates totals and tax and checks the mandatory fields of an invoice. For the IBAN it checks the check digits. If the read typically mixes up a zero and the letter O, the system proposes the corrected IBAN. The VAT ID is checked for format and structure.
It recognises an identical file as a duplicate. The same issuer and number with different content becomes a clarification case, so that no invoice is posted twice.
When master data matching is switched on, the system compares line items with the article master and sender and address with the partner master. Deviations in IBAN, VAT ID, address or phone become a note or a clarification case. The system never adopts suggestions on its own.
Clarifying
When a person has to decide
Whatever does not run fully automatically appears as a clarification case. It names the reason, the fields concerned and a suggestion. After your decision the document continues on its own. Confirmed fields are kept when a document is recognised again.
Empty, damaged or password-protected files are stored as records and appear as a clarification case together with the original file.
If something is missing from the supplier, a workflow prepares an enquiry as a draft that a person reviews and sends. Without administrator rights, recipients can be chosen only from addresses of the partner or its domains. Workflows are switched on per tenant.
Traceable
A source for every field, a log for every document
A value only counts as certain if the system finds it again in the document. The source names the page and position. A final step, the safety net, lowers fields when known patterns speak against them. It never raises them. A phone number mistaken for a document number is caught this way.
Whoever reviews sees the document and its fields side by side. Deviations are marked, and everything works from the keyboard. After approval the document is locked. Every later change is recorded in the audit log with its old and new value.
Document library
Further document types
Besides invoice, credit note, delivery note, contract, order confirmation, purchase order and quotation, Documents also recognises bank statements, timesheets and transport documents. For these document types, processing without human review is not yet released. A library adds payslip, official notice with a calculated deadline, certificate, reminder, termination letter and general correspondence.
Contracts have fields of their own per area, for example for rent, insurance, leasing or energy. The library and the contract areas are switched on per tenant. Only authorised people see payslips, and they run on the company's own infrastructure only.
Questions and answers
Questions about document recognition
Which file formats are read?
PDF, scans, photos, Word, Excel, OpenDocument, text, mails with attachments, ZIP archives and e-invoices as XML. The system looks inside the file and goes by its content, so a wrong extension does not matter.
What happens when the first and second read disagree?
A field on which the first and second read differ goes to arbitration. If it stays disputed, the document lands with a person for review. Nothing is posted until someone has decided.
Where can I see where a value comes from?
Next to every field you see the source with page and position in the document. In the review view the document and its fields sit side by side, and one click shows the spot in the image.
Can I submit documents through an interface?
Yes. Through the REST interface you submit files and fetch new, checked documents with a continuous pull. Each tenant has its own API key.