Macaly now runs its AI on its own servers in Europe
Since November 2026, Macaly has been running its AI models on its own servers in the EU. By default, users' prompts and their customers' data no longer end up in the US or China.

Václav Bešťák
Product Content Lead, Macaly

Prompts users send to Macaly are now processed on Macaly’s own servers in a European data center. The Macaly team built them in two months together with specialists from team.blue, the European group Macaly has been part of since December 2025.
That makes Macaly the only platform of its kind where AI models run entirely in the EU, on hardware it owns.
Independent of US tech giants
"With everything going on in the US, sovereignty is becoming a big deal," says Petr Brzek, CEO of Macaly. People and companies building websites and apps in Macaly hand AI models their code, and often their customers' data as well. They need to know where that data is processed.
“With everything going on in the US, sovereignty is becoming a big deal. That’s why we built our own servers in a European data center. Our users’ prompts and code don’t go to any model outside the EU.”
On its own servers, Macaly doesn't have to work around the usage limits set by large AI providers. Performance and availability are fully in its own hands.
"The servers are ours, and they run exactly the models we want. If the US government decided to restrict something, for example, we wouldn't be at the mercy of the biggest players," Brzek adds.
A faster, cheaper agent
DeepSeek is currently Macaly's default model. Because it runs on servers in the EU, prompts don't travel to the US or China. The same servers also run the latest model from French AI company Mistral.
In Macaly's internal benchmarks, DeepSeek works ten times faster than Claude Sonnet at the same level of quality. Users spend far less time waiting for each change to their project, and their credits stretch much further.
Users can still choose models from Anthropic or OpenAI, which run outside Macaly's servers. When picking a model, they can see right away which region it runs in.
Automated tests track the agent's quality
Macaly automatically checks whether the agent completes the tasks it's given. When the agent runs into a problem, an AI model proposes a fix. The fix is tested, and if it actually makes the agent better, it goes straight into production.
Since July 2026, this process has raised the agent's success rate from 94% to 99%. Failed tasks are now six times less common. "Six months ago, this wouldn't have been possible with open-weight models," says Brzek.
The whole project stays in the EU
Websites and apps built in Macaly have run on European servers for some time. Published projects are hosted in Germany, and user data is stored in Ireland.
With AI now running on Macaly's own infrastructure, the entire project can stay in the EU, from the first prompt to the published site. Users can tell their clients and customers exactly where their data is stored and processed.


