References
Who already uses it
DocAI is used by Hungarian companies in live operation.
DocAI is an enterprise AI system developed by K3Net Ltd. (Pécs, Hungary). It automatically reads incoming documents — invoices, contracts and vouchers — turns them into structured data and makes them searchable. DocAI runs on-premise: it operates on the customer's own server, and the documents never leave the organisation. The system was built for the Hungarian language, Hungarian regulation and NAV (tax authority) integration.
The key data in one place, in verifiable form. What is not listed here, we do not claim.
Last content review:
DocAI is developed by K3Net Ltd., a Hungarian software company based in Pécs. K3Net builds custom enterprise and industrial software: document management, production support, stock and asset inventory, working time registration and AI automation.
K3Net did not start out as an AI startup. We work on the internal systems of Hungarian mid-size and large companies — on financial, procurement and records management processes. DocAI grew out of that experience: not because a new technology became available, but because we saw the same problem at every client. Every morning, someone sits down and retypes data from a PDF into another system.
More about the company at k3.hu; the details of the operator of docai.hu are in the imprint.
Because at Hungarian companies document administration is the most expensive and least visible work. A mid-size company can receive several hundred incoming invoices, dozens of contracts and a large volume of vouchers every month — and someone retypes data from each of them by hand.
That work is costly three times over: it consumes expert time, it introduces errors, and it hides the knowledge inside the documents. The files sit on the server, but what they know — which contract expires when, which supplier raised prices, whether the same invoice arrived twice — is not accessible.
Our mission: turn business documents into knowledge while the data stays inside the company. AI should not be an external black box we pour our trade secrets into, but a tool running on our own server.
By default DocAI runs on-premise. Documents, extracted fields and the vector index all stay on the customer's own infrastructure. We do not send business documents to an external AI provider unless the customer explicitly chooses that model.
DocAI makes no automated business decisions. It extracts the data, recognises the document type and flags discrepancies — but approval, payment and signing remain human decisions. That is not a limitation but a design principle: responsibility cannot be delegated to a model.
Every AI answer carries its source: document chat shows which passage of which document it worked from. The processing chain — OCR, normalisation, classification, extraction — can be followed step by step.
Hungarian invoice formats, the NAV Online Invoice system, domestic records management rules and Hungarian language understanding are not plugins in the system. DocAI was built on them.
K3Net develops two products, and they are not peers. DocIT is the digital archive: it stores, organises and manages company documents, and handles approval workflows, invoicing and the NAV connection. DocAI is a standalone system that contains the same document management, with the intelligent layer built on top of it: it reads, understands and makes the content of those documents searchable.
| Aspect | DocIT | DocAI |
|---|---|---|
| What it is | Digital archive and document management system | Standalone document management system with an AI layer |
| Document management | Storage, organisation, approval workflows, invoicing, NAV submission | The same functions, as part of DocAI |
| AI capabilities | OCR | pro-OCR, data extraction, document type detection, semantic search, document chat |
| Question it answers | “Where is the document, and who approved it?” | “What is in the document, and what does it mean?” |
| Operation | SaaS, with a separate database per customer | On-premise or K3-managed |
| Website | docit.hu | docai.hu |
Both products are developed by K3Net Ltd. The records management domain expertise behind DocIT comes from its partner, DWS Digital Kft.
DocAI turns an incoming file into searchable knowledge in eight stages. With on-premise deployment every stage runs on the customer's own system.
Our model choices are settled by our own measurements rather than vendor marketing — on Hungarian documents, on our own hardware. The methodology and the raw results are public.
The measurement material behind our blog posts — configurations, scripts, results — is publicly available in the github.com/k3net/docai-evals repository.
Key information extraction is evaluated on our own Hungarian invoice corpus. The current production model reaches F1 = 0.983 on that benchmark.
We also publish what did not work out: NVFP4 quantisation gave +40% throughput, yet we stayed on FP8 because a third measurement showed a quality regression.
DocAI was built for organisations that receive many documents every day, and where the content of those documents is commercially sensitive.
Where DocAI is particularly warranted: if the organisation operates in a regulated industry, if the records contain trade secrets or personal data, or if internal policy prohibits the use of cloud-based AI services.
In three models, according to the customer's security and operational needs.
Data stays entirely on the customer's infrastructure.
In a Hungarian data centre, operated by K3Net.
Bringing in more capable models for more accurate results.
DocAI is designed to fit European and Hungarian regulatory expectations.
DocAI is designed as a decision-support system operating under human oversight: it makes no autonomous business decisions, its suggestions can be verified against the source documents, and the processing chain is documented step by step. Its design follows the EU AI Act principles of transparency, human oversight and documentability. How a specific deployment is classified under the AI Act depends on the intended purpose and the operating environment.
With on-premise deployment, personal data never leaves the customer's infrastructure, so there is no transfer to a third country. Access is role-based and operations are logged. The data processing of the docai.hu website itself is described in the privacy policy.
DocAI and DocIT connect to the NAV Online Invoice system, handle Hungarian invoice formats, and support domestic records management and archiving requirements.
DocAI is built by K3Net's own team in Pécs. Four competencies meet in it: software engineering (from the enterprise systems and integration side), AI infrastructure (model serving, quantisation, performance measurement on our own hardware), operations (on-premise and managed environments), and business process analysis (knowledge of financial, procurement and records management processes).
The same team works on DocIT, so the document management foundation and the AI layer do not come from separate worlds. We publish our measurements and what we learn while building — including our mistakes — on the blog.
DocAI is used by Hungarian companies in live operation.
DocAI is developed by K3Net Ltd. (K3Net Kft.), a Hungarian software company based in Pécs (Kétágú utca 7., 7634 Pécs, Hungary). K3Net builds custom enterprise and industrial software, handles system integration and document management, and is also the developer of the DocIT digital archive.
With on-premise deployment every document, extracted field and vector index stays on your own infrastructure; nothing is sent to an external provider. With K3-managed operation the data runs in a Hungarian data centre, operated by K3Net. The choice of model is yours.
On-premise operation means the entire system — OCR, data extraction, the search index, the vector database and the language model — runs on your own server rather than in an external cloud. Documents never leave the company network, and the system also works air-gapped, without an internet connection.
DocIT is a digital archive: it stores, organises and manages documents, and handles approval workflows, invoicing and the NAV (Hungarian tax authority) connection. DocAI is a standalone system that contains the same document management functions and adds an AI layer on top: it reads, understands and makes the content of those documents searchable. Both are developed by K3Net Ltd.
DocAI primarily processes invoices, contracts and accounting vouchers, but it can be applied to any text-based document. Scanned, paper-based records are read with OCR, then machine learning classifies the document type with a confidence threshold of 0.9–0.95.
DocAI is designed as a decision-support system under human oversight: it makes no autonomous business decisions, its suggestions can be verified against the source documents, and the processing chain is documented step by step. For GDPR, on-premise operation is the key advantage: there is no transfer to a third country, and access is role-based and logged. How a specific deployment is classified under the AI Act depends on the intended purpose and operating environment.
Yes. DocAI connects to the NAV Online Invoice system and can query authoritative company data from NAV, the Ministry of Justice (IM), VIES and KOMA sources. For existing ERP and accounting systems a custom integration is built, scoped separately for every deployment.
Yes, DocAI was built for Hungarian. Hungarian language understanding, Hungarian invoice formats and domestic records management rules are not features added later — they are the assumptions the system was designed around. The interface is available in both Hungarian and English.
Fill out the form below and our team will reach out shortly to schedule a demo.
We only use your data to respond to your inquiry. We do not share it with third parties. Legal basis: your consent and pre-contractual steps (GDPR Art. 6(1)(a) and (b)). For details, see our privacy policy.
