Skip to the collection

agent-native · Agent Runtime & Infrastructure · Verified 2026-09-20

Cua

Listed on The Agent-Native Index

Scale computer fleets for every agent

What it does

Cua gives computer-use agents real machines and a background driver. Cua Fleets keep warm pools of Linux, Windows, macOS, and Android desktops that code can claim for evals, RL, data generation, or batch rollouts. Cua Driver lets an agent click, type, scroll, read accessibility trees, and capture window state on macOS, Windows, and Linux without stealing the operator’s cursor or focus when the platform supports background delivery. Lume manages local Apple Silicon VMs. Cua Bench authors and scores computer-use tasks. CUA-S1 is a family of small specialist models for bounded GUI decisions (first profile: forms).

Computer-Use 2.0 in Cua’s docs treats the GUI as one tool surface beside code, shell, files, and APIs — not as a standalone “AI browser” product.


Agent-facing interfaces

How agents onboard

Interaction pattern: CLI + MCP (stdio) + Sandbox / Fleet SDK

# macOS / Linux
/bin/bash -c "$(curl -fsSL https://cua.ai/driver/install.sh)"

# Windows (PowerShell)
irm https://cua.ai/driver/install.ps1 | iex

cua-driver --version
cua-driver call list_apps
cua-driver mcp-config --client <client>

See the full onboarding section in the dossier.

Protocol surface

Interface Detail
Cua Driver CLI / MCP cua-driver — https://cua.ai/docs/how-to-guides/driver/install
Agent skill cua-driver skills install + embedded skill://cua-driver/
Sandbox SDK Python cua-sandbox — local and cloud runtimes
Fleet SDK TypeScript @trycua/fleet; cloud pools at https://run.cua.ai
Lume CLI https://cua.ai/lume/install.sh
Cua Bench uv tool install 'cua-bench[browser]'
Docs index https://cua.ai/docs

Verification

Last verified 2026-09-20.

Sources: