We write software. Mobile apps, web platforms, the dull reliable parts behind them. So the obvious question, asked by clients and by our own team, is why a studio in Belgrade spends its time on robots.
The short answer: in this region the machines arrived before the market around them did. The factories, the money and the rules are all moving. The part a buyer actually needs — somewhere to compare one machine against another — is not there yet. That part is software, and it is the part we took.
What arrived
A humanoid-robot factory is being opened in Šabac by AGIBOT together with Minth. AGIBOT is not a minor player — it ranked first in the world for humanoid shipments in 2025. Chinese commitments announced during the state visit spread across six Serbian cities, not one. And from January 2027 the EU Machinery Regulation 2023/1230 replaces the old Machinery Directive, which changes what a buyer has to demand from a supplier of autonomous equipment.
Those three facts have one thing in common: they are about supply. Factories, money, rules. None of them helps the company that has to decide which machine to buy — and that decision is where the money actually leaves the building.
What did not arrive
A neutral place to compare machines. Today that search ends in vendor PDFs, a sales call, and generated price pages whose numbers contradict each other and lead nowhere. The first honest question — who actually builds this robot, who services it here, and what happens when the vendor leaves the market — has no easy answer.
A robot sold under a European brand may be built elsewhere; a certificate may cover one radio module and not the machine. These are not conspiracies, they are ordinary facts of a young market. They just need somewhere to be written down.
The cost of that gap is not abstract. A facility manager who takes a number from a generated price page into a procurement meeting loses credibility the moment a supplier contradicts it. After that happens once, the whole category starts to look unserious — and a good machine loses a sale it deserved.
What we built
Rojium: a catalogue of 128 machines across 12 verticals, from commercial cleaning and mowers to prosthetics and robotaxis. Specifications come from the vendor with a named source. Prices stay request-only until regional terms are verified — we would rather show fewer numbers than invented ones.
It ships in English, Serbian Latin and Serbian Cyrillic, and the Cyrillic version is a first-class locale rather than a transliteration: the same machine card, the same filters and the same specification labels read correctly in all three. For a regional market that is the difference between a site that serves its audience and one that merely covers it.
Next to it sits a blog that explains the market instead of announcing it: what the rules of origin mean for a robot with a local badge, what a weapons policy on paper is worth, why a stock swing is not a buying signal.
What keeps a catalogue like this honest
Two rules, both of which cost us entries rather than win them.
Every claim carries a named source. If a figure cannot be traced back to the vendor's own documentation, it does not go on the card — even when a plausible number is easy to find elsewhere. And the price column stays empty until regional terms are verified, because a price without delivery, duties and service attached is not a price, it is a decoration.
A catalogue is only worth opening if a buyer can defend what they read in it in front of their own finance department. That is the whole standard, and it is also why the catalogue grew slower than it could have.
Why this is a software job
Because the missing part is not hardware. It is the layer that lets a buyer compare, verify and decide — a catalogue with sources, filters that match how people actually search, and three languages so that a Serbian company reads about a machine in its own.
It is also the kind of work we already do. The catalogue, the blog and the showcase run on one engine — the whitelabel core we develop in-house — so a new vertical, a new language or a new content block arrives as a configuration rather than a rebuild. That is what let the project start with research into real machines and their sources instead of with assembling a stack.
What we are watching next
The catalogue is the foundation, not the product. The work ahead is comparison: putting machines of one class side by side on the parameters a buyer actually decides on — cost of ownership, service reach, certification, and what happens when a vendor leaves the market.
We are not selling robots and we do not plan to. We build the infrastructure around them, which is exactly what a software studio should be doing while an industry is forming next door.
