- Documentor
- Features
What Documentor does.
From uploading a document to an answer that cites the exact page. Processing runs on its own, codes and labels sort the documents, search finds the passage and the chat answers from the passages it found.
Every document once, automatically.
Upload a PDF, a Word document, XPS, Markdown or an image through the web or copy it into the archive folder; anything that is not a PDF is converted automatically. No more clicking after that: the job queue takes the document through every phase, you see the progress as it goes, and a failure can be retried with one button. The same file is never processed twice.
- Conversion to PDFWord (DOC, DOCX, RTF), XPS, Markdown and images (TIFF, JPEG, PNG) are converted to PDF, the original file stays available for download
- Page sortinga page with a text layer, a scan, or a scan with an unreadable layer from the supplier – each takes its own route
- OCR of scansTesseract for Czech and English; pages lying on their side or upside down are rotated before recognition
- Page layouta layout model recognises headings, paragraphs, tables, images and charts even where OCR sees only letters
- AI descriptiona vision model only where needed: a table into a transcript, an image or chart into a description with the values read off, uncertain OCR once more
- PDF with texta scan gets an invisible text layer – the text can be searched and copied in the browser, the original stays as it was
- Annotationa main title, annotation and abstract for every document – in the list, in the detail and in search
- Optional stepsthe Archive administrator switches off what the archive does not need and switches on code reading or labelling; a diagram shows how the steps build on each other

A code on the page is data, not a picture.
QR Platba (the Czech QR payment code) on an invoice, a job barcode on a job ticket, a separator sheet in front of a batch of scans. During processing an archive can read the codes on every page, store them with the document and label documents according to the administrator’s rules. Reading needs no AI, and nothing leaves your infrastructure.
All the common ones
QR, Data Matrix, Aztec, PDF417, EAN, UPC, Code 128, Code 39, Code 93, ITF and Codabar – even rotated sideways or away from the centre of the page. Reading is optional in each archive and switched off at first.
Search by code
The payment reference (variable symbol) from a QR Platba code or a job number from a barcode finds the document in the document list. The detail shows the codes on the page and frames them on the page image; the code text can be copied.
A label by pattern
The Archive administrator writes a pattern (a regular expression), optionally a code type, and a label. Test shows how many documents would get the label; after saving, the labels are recalculated without reading the pages again.
| I want a label for | Pattern | Code type |
|---|---|---|
| invoices with QR Platba | ^SPD\* | QR code |
| payments to a specific account | ACC:CZ6508000000192000145399 | QR code |
| goods with a thirteen-digit code | ^\d{13}$ | EAN-13 |
| separator sheet “ARCHIV/ZOO” | ^ARCHIV/ZOO/ | Code 128 |

Invoice, contract, report – at a glance.
A label is a short tag by which you divide up and find the documents in an archive. The document list has a filter by label with counts, and the document detail shows where a label came from and how confident whoever assigned it was.
- Labels from a language model
- Administrator’s listthe Archive administrator writes the labels, each with a description of how to recognise the document; the model chooses only from them
- Confidence thresholda label below the threshold is not assigned; the number of labels per document depends on whether labels are combined
- Whole archiveafter the list changes, documents already in the archive can be labelled again too
- Costone model call per document; an external model only in an archive that has allowed it
- Labels from codes
- No AIa document gets the label when its QR code or barcode matches the rule’s pattern
- Immediatelya change to the rules applies to the whole archive once saved, pages are not read again
- Side by sidelabels from codes and from the model look the same, the tooltip shows their source
- Sortingcodes can also send an incoming document to an archive – see Document sorting
It finds the passage, not just the document.
Search combines meaning (vectors) with exact words (BM25), and a reranker sorts the hits. The result is a coherent passage with its heading and a preview of the page with the passage highlighted – one click takes you exactly there in the document.

“Voltage drop” also finds “fall in voltage”
A dense vector understands the meaning, BM25 keeps exact designations and numbers, and a reranker puts the order together. The same text from duplicate documents is not repeated, and for a query the archive has nothing on, search says so: Nothing found.
A table knows what it is about
Tables, images and charts go into the index with the AI description, their caption and the introductory sentence from the document – even the continuation of a table on the next page carries the title of its first part. You can search tables only, or images and charts only.
One query, all your archives
Pick the archives you have access to and they are searched at once. Each archive has its own index, results are grouped by document, and a relevance threshold hides less relevant hits. The Exact text mode finds designations and numbers literally.
Every statement comes with the page it is from.
The chat answers only from what is actually in the archive and adds a source number to every statement. Click it to see a crop of the page with the passage framed. What the archive does not contain, the AI does not make up.

- Chat
- Cited passagesa source number for every statement, a preview of the cited passage in a panel next to the answer
- Artefactsa table, image or chart from the answer opens in the panel; a chart can also be redrawn from the values the AI read off
- In-depth readingbefore answering, the AI reads around the passages it found, the neighbouring pages and the tables next to them
- Conversationssaved and private, follow-up questions, a Think switch with the model’s reasoning shown live
- Research
- Follows references“the cross-section is given in table D.1 in the appendix”, “see another standard” – research finds and reads the table and the standard on its own
- Overview questions“which standards deal with emergency lighting” – it goes through the whole archive, document by document
- Live progressyou see what the AI is searching and reading right now; Answer now or Stop at any time
- Depthstandard, thorough or exhaustive – depending on how general the question is
- ces · engtext recognition of scans, rotated pages straighten themselves
- PDF + textevery scan gets a text layer, the original stays unchanged
- [1] … [n]every statement in an answer points to a specific page of the document
You know what is going on, even when you are not looking.
Processing runs unattended, but not silently: you see the state of the archive, the queue and the history in the application, and you get an e-mail about whatever concerns you.
Queue and log
The archive overview shows documents by state, the queue shows jobs by phase, and the log shows what the workers processed and what ended with an error – when, which document and which step. A failed document can be processed again or deleted.
An e-mail when there is something to do
Processing errors, new documents, requests awaiting approval and incoming documents to sort. Everyone chooses immediately, daily digest or off; a group administrator sets the default for the members.
Explanations where you need them
Whatever is not obvious at first sight is explained by an (i) icon right in the application. The manual is in the application, and after every new version a window with what is new appears once.
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