Cua
Cua provides open-source infrastructure for running AI agents that can see a screen, reason about it, and act.
Our verdict
Who it's for
Run AI agents that see, reason, and act on any desktop
Who should look elsewhere
Cloud service requires booking a call for access, not self-serve
Pros
- +Open-source and self-hostable, giving full control over infrastructure
- +Supports all major desktop operating systems plus Android
- +Built by researchers with deep expertise in computer-use agents (Windows Agent Arena)
Cons
- −Cloud service requires booking a call for access, not self-serve
- −Primarily designed for developers and researchers, not end-users
Key features
- ✓Multi-OS desktop environments for agents (Linux, Windows, macOS, Android)
- ✓Lume: open-source CLI for running native macOS VMs on Apple Silicon
- ✓Cua Cloud: managed cloud sandboxes for scaling agent fleets
- ✓Agent SDK for building and orchestrating computer-use agents
- ✓Cua-Bench: benchmark suite for evaluating computer-use agents on desktop and mobile tasks
- ✓VLM Router and Playground for vision-language model integration
- ✓CuaBot: multi-player computer-use for coding agents
- ✓Self-hostable, MIT-licensed open-source components
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Local | — | month | Run GUI Docker; QEMU; Apple VZ sandboxes locally |
| Dedicated fleets | — | month | Hosted; BYOC; On-prem infrastructure |