← Back to open sourceFeatured starting point
Choose the component that matches your computer-use problem before installing a fleet of tools.
Cua explained: computers, drivers and benchmarks for agentsCuatrycua/cua
Start reading →Suggested learning path
01 → 09
01·3 min read·Search intent: Learning
Choose the component that matches your computer-use problem before installing a fleet of tools.
Cua explained: computers, drivers and benchmarks for agentsCuatrycua/cua
·3 min read·Search intent: Learning
Use the documented Calculator exercise as a bounded first result.
Cua quickstart: verify one desktop action safelyCuatrycua/cua
·3 min read·Search intent: Learning
Choose isolation and lifecycle before exposing any computer-use tool.
Deploy Cua: local sandbox, host driver or cloud FleetCuatrycua/cua
·3 min read·Search intent: Learning
Map the shared guest state and the separate authorization boundary.
Cua architecture: code and GUI on one isolated machineCuatrycua/cua
·3 min read·Search intent: Learning
Use the wrapper and fleet modules as reading anchors, without claiming a whole-monorepo audit.
Cua source code: trace a Python Driver call to the fleet boundaryCuatrycua/cua
·3 min read·Search intent: Learning
Define task success from the environment, not from the agent’s own report.
Evaluate a computer-use agent with Cua BenchCuatrycua/cua
·3 min read·Search intent: Learning
Apply runtime controls before giving an agent desktop or shell access.
Cua security: policy ceilings, host permissions and screenshotsCuatrycua/cua
·3 min read·Search intent: Learning
Use the narrowest component that can produce a verifiable result.
Choose Cua components for your computer-use workloadCuatrycua/cua
·3 min read·Search intent: Learning
Capture observation, action, policy and final state for one bounded desktop task.
Build a Cua task receipt that proves what the agent changedCuatrycua/cua
