About
Built by a company that already ran its own infrastructure.
Zarx is a service of NPAW. This is where it comes from, what we believe, who is building it, and how to reach us.
The story
From video analytics to sovereign inference.
NPAW (NicePeopleAtWork) is a video analytics company founded in Barcelona in 2008. It has run its own data infrastructure for streaming services since then. Streaming services in Europe and beyond send it their playback data; NPAW turns it into the numbers those services run on. Doing that for years meant running a large data platform ourselves, in our own racks, on our own network.
When language models became useful for our products, we did what we had always done: we put them on our own servers. We bought datacenter GPUs, ran open-weight models with vLLM and SGLang, and built an OpenAI-compatible gateway so our teams could use them like any other API. That gateway now serves NPAW's products in production.
Then customers started asking the question we had already answered for ourselves: where does the prompt go? Banks, telcos and public bodies in Europe want inference they can place on a map and under a law. Zarx is that gateway, on that iron, opened to them.
NPAW's platform processes more than two trillion events a day for over 200 streaming services in 197 countries, up to 15 billion video plays a month.
What we believe
Three principles we will not trade.
Own the iron.
We buy and operate the servers. Renting capacity from a hyperscaler would make the jurisdiction claim a contract clause; owning it makes the claim a fact you can visit.
Serve open weights.
A model we cannot download is a model whose provider is in the data path. Every model on Zarx has published weights and a license we can read.
Publish what we can prove.
No performance, price or comparison figure without evidence you can check. Every statement on this site traces to a source, and owner-only facts are marked until confirmed.
Team
The people who run the platform.
The team behind Zarx built and operates NPAW's inference gateway, its GPU servers and the network they sit in. Operating models in production, on hardware we own, for products with real users, is what we do every day. The gateway behind Zarx is the one NPAW runs in production for its own products, serving its own models and third-party providers through one API.
Zarx is led by Ferran Gutiérrez Vilaró, CEO of NPAW, and built by NPAW's engineering and AI teams in Barcelona, the people who run NPAW's inference gateway in production.
By NPAW
What NPAW brings to Zarx.
Operational history. NPAW has run data infrastructure for demanding customers since 2008. Inference is a new workload on a familiar discipline: capacity, observability, on-call, security reviews with enterprise buyers.
A European home. NPAW is a Spanish company headquartered in Barcelona. Its legal entity, its servers and its network allocation are in Spain, and the service is operated from there.
Skin in the game. The same platform serves NPAW's own products. Reliability problems reach us before they reach you.
Talk to the people who run it.
Early access is a small program by design. Tell us what you run inference on today and what has to stay in your jurisdiction.