Your staff already use ChatGPT — with company documents, without oversight. Banning it does not work; leaving it unregulated is a risk. You need an internal alternative that is better than what they are using in secret.
The AI project is ready, everyone is enthusiastic, and then the data protection officer asks the question: where does the data go? And the project stops. For us this is where it starts: on-premise and hybrid architectures take AI into places cloud vendors walk away from.
Thousands of documents, three systems, and know-how in the heads of colleagues approaching retirement. A permission-aware knowledge base agent answers in minutes from material that takes half a day to track down today.
Quote requests, invoices, email triage, reconciling data across systems. These processes are candidates for agent automation — with people deciding and AI doing the work.
We deliver this two ways: built on our own Athene AI platform (faster, with predictable licence costs), or as fully custom development if the process calls for it. The two do not compete: the assessment decides which is worth it for you.
Open-weight language models (Llama, Mistral) on the company's own GPU server, in a private cloud or in a hybrid setup — in production, not in a lab. Few vendors on the Hungarian market can demonstrate this today; most "enterprise AI" is in fact a front end on a US API.
We tune the models to Hungarian enterprise language — the agent does not answer your colleagues in translation-flavoured Hungarian.
Risk classification, transparency elements, data processing agreement — part of the deployment, not an extra charge. Details on the Security and compliance page.
An average of 10+ years of development experience, and your project does not disappear into an account hierarchy: you talk to the decision-maker, not to a ticketing system.
Processes, data assets, infrastructure, compliance constraints. It ends with a written proposal: what is worth handing to agents, on what architecture, at what cost.
A working system on real data, with success criteria agreed in advance and a go/no-go decision point. The usual pilot on the market takes 90 days — we deliver in half that time.
Gradual expansion, training, operations backed by an SLA. The system stays with you — together with the source code and the documentation.
.svg){kind=small}