Scan, Sign & AI Chat
Mobile App
The mobile app pairs with your account via QR code and brings scanner, signature, chat and documents to your phone.
Four special cases, four solutions: the Split Agent separates a batch scan into individual documents, PaperOffice reads handwriting, the Trust Agent finds covered-up bank details, and the Redact Agent pseudonymizes a payslip.
Hello and welcome to the PaperOffice AI Academy. Today, we will discuss four specialists in PaperOffice that handle tedious special cases for you. First, the result: five clean documents with descriptive names are extracted from a seven-page batch scan. A proof of delivery with handwritten notes is read. An invoice where someone has obscured the bank details is immediately flagged. And a payslip is sent to the tax advisor in pseudonymized form. We will look at where these agents are located in PaperOffice, and then examine each one individually: Split, handwriting recognition, Trust, and Redact.
You can find the special agents on the left under Document AI: Split Agent, Trust Agent and Redact Agent. This is where you create agents with fixed settings if you need to perform the same task repeatedly. The process is initiated from the document list: select documents, Actions, Agents and workflows. Here you will find Split, Trust and Redact. Thus, a document is always selected first, followed by the agent.
First issue: In the mailroom, an entire stack is scanned in one go. This results in a seven-page PDF containing a purchase order, delivery note, invoice, and more, including a separator sheet. For this purpose, we create a Split Agent. We name it Split mailroom and select the level Premium. As a result, the individual parts will be assigned not only numbers but also the document type, date, and sender in their names. In the Split method tab, you define how the agent recognizes where a new document begins: automatically, by a cover page, by a separator code, by a change in document type, after a fixed number of pages, per person or file, or with your own rule. We leave it set to Automatic. Then PaperOffice detects the boundaries based on content, sender, date, and document type; separator sheets are possible but not required. As the target, we select the existing Workspace Mailroom 2026. Now for the batch scan: select the items in the list, then click Actions, Split and Run via Split Agent. Click on Start, and the agent will handle the rest in the background. After about a minute, the monitor displays the created documents: five documents have been generated from seven pages, each with its document type, date, and sender in the filename. The separator sheet has been removed, and the original scan is preserved. A quick glance at the list is still worthwhile: If the separation is correct, you can proceed as usual with data fields, workflows, and everything else. A look at one of the new documents, the order to Veldtmann: PaperOffice has extracted it like any other document, including title and a summary with amount.
Second issue: handwriting. The customer confirms our delivery on the proof of delivery document by hand: 98 packages received, box 7 damaged, 2 packages rejected. A scan is returned. For this, you do not need your own agent: Starting from level AI-DMS Ultra, PaperOffice recognizes handwriting during processing automatically. We therefore load the scan into a Workspace with Ultra. Even the summary mentions it: 98 out of 100 packages arrived, box 7 damaged. And in a separate section, Handwritten notes, you will find everything that was entered manually, up to the note about the complaint. The stamp is also recognized by PaperOffice. This way, you can immediately see that a complaint is pending. Please briefly verify any handwritten information on the document itself, especially numbers.
Third issue: Two invoices from Kaltwerk with the same number. At first glance they appear identical, but in one case someone has covered the bank details at the bottom with a white box and written new information over it. The Trust Agent checks whether any changes have been made to a PDF. We mark the copy, select Trust, and start the integrity check. For the document type, we choose Invoice, so that PaperOffice additionally pays attention to typical patterns in invoices. With Premium, PaperOffice highlights the findings directly in the document and creates a Marked evidence PDF. The result: Review recommended, with three findings. The most significant one: In the text of the PDF, a bank account number is present, which is obscured in the visible document by a shape drawn later, and a different one is displayed there. Additionally, the invoice lists bank account numbers from various countries. The risk score also appears as an icon in the list. The review by PaperOffice covered four areas: hidden content that is overlooked during normal viewing, the technical structure of the PDF, a comparison between what is visible and what is stored in the document, and additional patterns for invoices using the invoice profile. For comparison, the original: no anomalies detected. The Trust Agent does not tell you who committed fraud, but it shows you where to look more closely before making a payment. Another example, this time with the profile Application and CV: This resume contains an instruction to AI systems in very bright and tiny font, practically invisible to humans. The Trust Agent detects it and flags the resume as suspicious.
Fourth issue: The tax advisor requires a payslip but not the employee’s name, address, or bank details. We mark the payroll statement and select Redact, Edit with preview. First, the scenario, i.e., what is being redacted: for example, for a GDPR disclosure to data subjects, for sharing with business partners and service providers, or only for financial or contact details. PaperOffice sets the corresponding rules accordingly. For the payslip, we select Personnel File. During fine-tuning, we specify what should remain unchanged: the company name Northstar. For the output, we select Pseudonymize, which is available with the model Ultra. Before anything happens, the AI redaction review highlights each field that is being replaced. If something doesn’t look right, you can toggle it with a single click, and simply drag in any missing fields. Afterwards, the copy displays a fictitious name instead of the real one, and an alternative address in place of the actual one. This keeps the document readable while concealing the true names. Health data and signatures remain blacked out. The table below indicates which name corresponds to which entry. This information is only available in this dialog and never in the PDF file. Furthermore, new pseudonyms are assigned with each run. Please copy them if you need to maintain the mapping. The original remains unchanged, and the pseudonymized copy is placed directly beside it. Please take a moment to review it before sending. In summary: The Split Agent splits batch scans into individual documents. Handwriting is automatically recognized by PaperOffice in Workspaces with AI-DMS Ultra. The Trust Agent indicates where a PDF has been altered. And the Redact Agent redacts or pseudonymizes, with review before applying. It is always launched from the document list, via Actions and Agents and workflows. Until next time in the PaperOffice AI Academy.
AI-DMS
Documents. Intelligence. Real impact.
PaperOffice AI is the AI-DMS. Four functional areas on one DMS foundation, with three ways in.
Features, modules and benefits at a glance.
Capture, organize, search and collaborate on documents.
Plug into your systems and AI workflows.
No installation, ready in seconds.
Search all tools