Cua is unclaimed —

Cua

Scale computer fleets for computer-use agents·cua.ai

Free plan · Open source See pricing ↓

Cua is open-source infrastructure and a cloud platform for computer-use AI agents. It provides an open-source MCP/CLI driver (Cua Driver) that lets agents click, type, and inspect native desktop apps without taking over the cursor, plus Cua Sandbox and Cua Fleets for provisioning disposable or pooled GUI environments across Linux, Windows, macOS, and Android. It also offers Cua Bench for benchmarking computer-use agents and verified trajectory datasets for training and evaluation.

What it's for

Computer-Use Agent Development: build and test agents that interact with GUIsAutonomous Agent Deployment: deploy production computer-use agents at scaleSecure Agent Sandboxing: test agents in isolated environments before productionMulti-Agent Systems: run multiple computer-use agents in separate secure sandboxesAgent Research: experiment with computer-use capabilities in controlled environmentsEnterprise Agent Automation: deploy secure, managed agent infrastructure for teamsRunning computer-use agent training, eval, and data-generation workloads in parallel

Features 20

API keys for programmatic agent deployment and management
Auto-scaling and resource management for production agent systems
Background computer-use driver for native desktop apps (click, type, scroll)
Batch throughput via pre-booted warm pools that scale to zero when idle
Cross-OS fleets: one API boots Linux, Windows, macOS, and Android machines
Cua Bench: eval and gym authoring for cross-surface agent benchmarks
Cua Fleets: elastic infrastructure for coordinated sandbox pools for evals and RL loops
Docker-based container support for agent workflows
Enterprise SSO via Clerk for team-based agent management
Human-reviewed golden trajectories and step-level annotations
Inspects accessibility trees and captures window state without stealing cursor/focus
Local runtimes (Docker, QEMU, Apple VZ) or Cua Cloud sandbox provisioning
Lume: macOS VM management on Apple Silicon using Virtualization.framework
MCP server and CLI interface for agents and humans
Multiple container sizes (small, medium, large, xlarge)
Real-time credit-based billing
Sandbox playground for testing computer-use agents
SSL certificate management for secure agent communication
Unified model provider access to 100+ VLMs
Verified trajectory datasets for training and evaluation

At a glance

free tieropen sourceself-hostableAPI

TypeWeb app

DeploymentHybrid

PlatformsWeb, Windows, macOS, Linux, Android

ForAI/ML researchers building computer-use agents, developers creating autonomous agents that control computers, companies deploying production computer-use agent systems, startups building agent-powered automation products, security teams testing agents in isolated sandbox environments, students and educators learning about computer-use AI

CompanyCua AI, Inc. · 2025

Integrations

H CompanyDroidCodeRabbitFactoryQwen CodeClickyClaude DesktopOllamaModel Context ProtocolAnthropic ClaudeHermes

Complianceself-reported

HIPAASOC 2 Type IISOC 2 Type I

Resources

Pricing

Pro plan is pay-as-you-go credit-based (minimum $10 purchase, up to $5000), not a fixed monthly fee. Enterprise pricing is custom/quote-based.

Open SourceFree

For developers and open-source projects who prefer self-hosting.

Self-hosted, local environments only

ProQuote

Most Popular; pay-as-you-go credit-based pricing, starting at $10

Minimum $10 credit purchase, up to $5000

EnterpriseQuote

Custom pricing; scalable cloud sandboxes for large teams and organizations, book a demo

Cua