Your document archives,
under control.
Documentor takes your PDFs – scans as well as documents with text – and recognises the text, tables, images and charts in them, indexes them and answers your question with a citation of the exact page. Archives belong to user groups, search runs across them, and AI answers only from your documents. It runs in the browser, nothing to install.
- PDF, Word, XPS, images
- OCR of scans
- Tables, images, charts
- QR codes and barcodes
- Search across archives
- Answers with citations
- Research
- Document intake and triage
Drop in a PDF. The rest is automatic.
Standards, equipment records, drawings, reports, contracts – scanned twenty years ago or issued yesterday. Documentor processes them once, keeps the original and a PDF with a text layer, and from then on you search them and ask questions.
A readable document from a scan
Pages are sorted, scans go through text and layout recognition, and AI describes tables, images and charts. The result is a PDF whose text you can search and copy, plus an index for searching.
- OCR for Czech and English, rotated scans are straightened
- Headings, paragraphs, tables, images and charts as blocks
- QR codes and barcodes, labels, an annotation and abstract
By meaning and by words
Search finds the passage even when it is worded differently from your query, and also matches exact designations (“ČSN 33 2000-5-52”, “table D.1”). The result shows the page with the passage highlighted.
- Several archives at once, results grouped by document
- Tables only, images and charts only, a relevance threshold
- AI descriptions of tables and images are searched too
An answer with a citation, not an impression
The AI answers in Czech using only the documents it found, and backs every statement with a link to the exact place. Research looks up on its own the table in an appendix or the standard that the text only refers to.
- Preview of the cited passage next to the answer
- A chart from the document redrawn from its values
- Saved conversations, research depth
An archive for a team, a department or a group of companies.
Who may read what, where files are stored, which AI models are used and where a document that has just arrived by e-mail belongs – Documentor takes care of that for you. See how.
Everyone sees what they should
Archives belong to user groups, and roles apply to individual people as well as whole groups. A group has its own storage and AI models, and usage is counted by person, group and archive.
Documents find their archive
From a folder, another system or e-mail. The archive is suggested by the sender, a QR code or barcode, similarity to the content of the archives or a language model; a sorter confirms.
Standards, building records, invoices
Eight typical kinds of documentation – what Documentor does with them and what people ask in the chat.
What it could look like for you
An engineering plant, a town’s building management, a design studio and a group of companies: fictional organisations with the full setup.
Why not a shared drive and a general chatbot.
Three things a “folder of PDFs with full-text search” does not have: scans and tables as real content, answers backed by a page of the document, and search across separate archives without mixing their data.
Scans are text
Twenty-year-old scans of standards and equipment records get a text layer and an index. An unreadable layer from a converter is detected and replaced with our own recognition.
Tables and charts are not blind spots
Full-text search sees scattered numbers in a table. Documentor has it transcribed with its caption and conditions, and a chart with its values read off – and the AI answers using them.
An answer with a source, not an impression
Every statement points to a page you can check with one click. The AI must not fill in from its own memory, and it says so when a question falls outside the archive.
Your data stays with you
The index, embeddings and reranker run on servers you choose – yours or ours, never a third party’s. An external model gets documents only from an archive that has allowed it; from the others, never.
Once and automatically
Processing runs in a queue without supervision, you see its status and errors, and a retry is one button. A change in the archive settings is applied by reindexing, not by hand.
Help where you need it
Whatever is not obvious at first sight is explained by an (i) icon right in the application: processing phases, document states, search options, recognition confidence. What is new in each version, inside the application.
From your first PDFs to answers with citations
-
01
We set up archives and groups
Archives by type of documentation, groups by who works with it. For each archive: roles, storage, AI models and the choice whether external AI is allowed.
-
02
You upload documents
PDF, Word, XPS, Markdown or images through the web or by copying them into the archive folder, or another system or e-mail sends them in. Hundreds of documents at once.
-
03
Processing runs on its own
The queue takes every document through page sorting, OCR, layout, AI description and indexing, optionally also code reading and labelling. You see the status and errors in an overview.
-
04
You search and ask
Search with a preview of the passage, chat with citations, research across the whole archive. From the result straight into the document.
-
05
You fine-tune and expand
Instructions for the AI, labels, sorting rules, more archives and groups. Updates, backups and monitoring are on us.
A service in the browser, the data where you want it.
Documentor runs in containers: the web, processing workers, database, index and storage. Users install nothing, a browser is enough. You choose where each part runs: entirely on your premises, your data with us running the rest, or entirely with us.
- Chrome · Edge · Safari
- PostgreSQL
- Qdrant
- S3 · SeaweedFS
- Tesseract
- Kubernetes
- Ollama
- Application
- web-based, Blazor, .NET 10
- Sign-in
- Microsoft Entra ID (organisation and personal accounts) and Google
- Archives
- several in one instance, owned by groups, each with its own index and storage
- Formats
- PDF, Word (DOC, DOCX, RTF), XPS, Markdown and images (TIFF, JPEG, PNG), all converted to PDF
- Roles
- Reader, Processor and Archive administrator; Archive creator, group administrator, Supervisor
- Data
- PostgreSQL and a Qdrant index; originals and pages in S3-compatible storage (SeaweedFS in our deployments)
- OCR and codes
- Tesseract 5, a page layout model and reading of QR codes and barcodes, no cloud
- AI
- models on your own server (Ollama) set per group; external ones only for an archive that allows it
- Processing
- job queue, several workers, retry after an error
- Appearance
- light and dark themes, help, a manual and What’s new in the application
Everything inside your network
The application, processing, database, index and storage run on your server or in your cluster. Documents never leave your infrastructure; your team runs it, or we do remotely.
- Your own S3 storage, database and index
- AI models on your server
- Updates as agreed
Your files stay with you
Originals, page images and PDFs with text are kept in your S3-compatible storage. We run the application, processing, the database with the recognised text and the index.
- Storage set per group
- Each group can have its own
- Updates and monitoring are on us
You just sign in
We run all of Documentor in our cluster, with archives kept separate by group. No infrastructure of your own; updates, backups and monitoring are on us.
- No servers on your side
- Backups and monitoring
- External AI only with the archive’s consent
No extra infrastructure licences
Deployment relies only on open-source components: Kubernetes (k3s in our case), PostgreSQL, the Qdrant vector index, S3-compatible storage (SeaweedFS in our case), Tesseract OCR and AI models via Ollama. You buy no extra licences for the database, storage or orchestration.
Encryption, backups and your own storage
Archive files encrypted with the archive’s own key, encrypted backups to separate storage and storage you choose yourself. External AI only with the archive’s consent.
How much does it cost?
It depends mainly on the deployment option and on the volume of data – how many documents and pages the archives hold and how fast they grow. We therefore put the price together for each customer, from three parts.
The application itself
Documentor as such: document processing, search, chat and research, archives and groups, updates and support. The price depends on the deployment option and on the volume of data.
- On your premises, your data with us running it, or entirely with us
- By the number and size of archives
- Updates and support
Integration and initial conversion
One-off: setting up archives and groups, connecting sign-in and document intake from e-mail, a scanner or another system, and if needed the initial conversion of your existing documents into the archives.
- Sign-in with Microsoft or Google accounts
- Intake from e-mail, an S3 folder or an API
- Conversion of existing documents, if needed
Costs of AI models
For smaller deployments that we run for you, AI is included: the price covers a monthly volume of questions. Larger customers and deployments on your premises take care of the models themselves – they pay the provider directly, or run their own on their server. The costs then depend on how many questions you ask, how many research runs you start and how many pages with images and tables you process.
- Smaller deployments with us: AI included, with a monthly volume of questions
- Larger deployments: your own key to an external model, or your own model on your server
- The application shows usage by people, groups and archives
Tell us how many documents you have and how you want to deploy – we will prepare an offer →
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