A Story of Computer-Use: Where We Started, Where We're Headed
From OpenAI Universe in 2016 to Clawdbot's virality - the timeline of computer-use agents converging now.

Cua is a cloud desktop platform that serves computer-use agents.
cua.aiCrawled Oct 16, 2026
From OpenAI Universe in 2016 to Clawdbot's virality - the timeline of computer-use agents converging now.
CuaBot brings multi-player computer-use to any coding agent. Run sandboxed Linux desktops with agent and human cursors coexisting on the same screen.
Create focused Computer sessions limited to specific apps. Enable single-agent focus on iPhone automation or run multiple agents in parallel controlling different applications.
News and updates from the Cua team on cloud desktops, AI agents, and computer-use infrastructure.
The page describes building computer-use agents with JavaScript and TypeScript using OpenAI's computer-use-preview model and the @trycua/computer library, including automating tasks in cloud containers.
The page provides a step-by-step guide to building an Operator on macOS.
The page describes the cua-agent framework for building computer-use agents with multiple model providers and a simplified API.
Announcing general availability of Cloud Windows Sandboxes and preview of macOS Sandboxes on Cua Cloud.
The Cua team describes using computer-use agents for growth hacking at Toronto Devfest.
Two-track competition for SOTA Computer-Use Agents: on-site Hack the North participants compete via OSWorld benchmark, while global participants build with Cua + Ollama.
Benchmark GUI-capable agents on hundreds of Computer Use tasks with ComputerAgent and HUD integration.
Cua Playground lets users try computer-use agents and cloud sandboxes in a browser.
The Cua VLM Router lets users switch between different computer-use model providers.
Cua tested Gemini 3.5 Flash's native Computer Use tool on Cua-Bench's KiCad EDA suite and reported a mean reward of 0.267.
The page discusses Apple's Containerization framework announcement.
Cua Driver lets agents control Linux desktop apps in the background on X11 and XWayland. It uses AT-SPI 2 over D-Bus, XTEST, and a separate agent cursor.
The page describes an open-source macOS driver that lets agents drive Mac apps in the background without moving the cursor or stealing focus, and explains the SkyLight internals behind it.
Cua Driver now runs Windows apps in the background for any agent loop. One driver across Win32, WPF, WinUI, UWP, Electron, and legacy controls, handing the model window pixels, accessibility trees, and an action layer.
Deploy Computer-Use Agents at scale with Cua Cloud Containers. Launch pre-configured cloud environments with all agent frameworks ready to go.
The Cua CLI lets users manage Cua Cloud resources from a terminal.
A roundup of the most exciting computer-use agent research presented at NeurIPS 2025.
The page explains how the @sandboxed decorator runs Python functions in isolated virtual environments within Cua containers.
The page explains how to use CUA's Gradio UI to create human trajectories from computer interactions and share them on Hugging Face.
The cua Trajectory Viewer lets users explore, debug, and analyze agent actions by stepping through observations and actions in a browser.
Cua offers Ubuntu Docker support, providing a Linux desktop in a Docker container that can be viewed in a browser without a VM or extra clients. It works on macOS, Windows, and Linux.
The Cua team sponsors Hack the North and shares its experience with hackathon judging.
Sometimes the best AI agent is a human.
Windows Sandbox is a lightweight, disposable virtual machine built into every Windows PC - perfect for safely testing computer-use agents.
Cua builds cloud desktops for AI agents. The page describes Cua's story, team, and principles.
Scale Linux, Windows, macOS, and Android computer fleets for computer-use agents with one open-source MCP/CLI driver.
The page describes how Cua collects, uses, and protects personal data when people use cua.ai.