Cua Cloud
Cloud desktops for AI agents
Linux, Windows, macOS, Android.
Philosophy
How we think about building Cua.
The principles behind our product, research, and engineering decisions.
01
Agents need real computers, on every OS.
We believe there should be one protocol and one SDK to access any operating system. Linux, Windows, macOS, Android. Agents that do real work need a display, a file system, and installed software on a full OS, and the interface to all of them should be the same.
02
Research shapes what we build.
We started by publishing Windows Agent Arena at ICML, then built Cua to fill the gaps that paper exposed. Cua-Bench came from watching teams struggle to evaluate their agents on real desktops. We do the research first, find what's missing, then build it.
03
Open source as transparency.
Lume, the Agent SDK, and CuaBench are open source because infrastructure you depend on should be inspectable and forkable. It also means bugs get found faster, integrations get built by the people who need them, and we can't hide behind a sales process when something breaks.
04
One API call to a running desktop.
If it takes more than a few seconds to spin up a sandbox, that's a bug. If you need to talk to sales to get started, that's a bigger bug. We optimize for the shortest path between having an idea and watching an agent execute it.
05
The infrastructure should disappear.
You shouldn't think about VM provisioning, display protocols, or certificate management. You should think about what your agent does. Every layer of complexity we absorb is a layer your team doesn't have to staff for.
06
We run on the same infrastructure you do.
Cua uses Cua. Our internal agents, our benchmarks, our demos all run on the same platform we deploy to enterprises and research labs. If a sandbox is slow to create or a screenshot takes too long, we feel it before you do.
07
Build for the teams shipping today.
Computer-use agents are in production right now. We talk to the people running them, find out where they get stuck, and fix that. Our roadmap comes from their real problems, not from imagining what someone might need in two years.