agent-cu

Overview

agent-cu is a command-line control layer for desktop applications, using clicks, typing and other actions directed through the accessibility tree. It suits developers, AI-agent users and technically minded operators who want local automation across major desktop systems. Its broad action set and agent connections are useful, but its discontinued status is a serious qualification for any new deployment.

Key features

Instead of depending only on screen coordinates, agent-cu reads the accessibility tree used by screen readers and acts on referenced interface elements. That makes snapshots, selectors, clicks, typing, scrolling, dragging and window management practical from a terminal, while JSON output supports scripts and agent pipelines. It is a strong fit for structured, visible desktop tasks; it is not a screen-understanding system, so it cannot interpret pixels the way a visual computer-use tool does.

Batched commands and reusable YAML workflows turn individual actions into repeatable sequences. Electron support through CDP covers Slack, Cursor, VS Code, Postman, Discord and Notion, extending the tool beyond conventional native windows. The project describes skills for Claude Code, Cursor, Codex, Copilot, OpenCode, Cline and more than 40 other agents, which is valuable when an existing agent needs a local desktop-control backend.

Browser control is supported, but the deployment remains local. The tool requires explicit operating-system accessibility permission to read and control applications. Its security policy states that it does not transmit data over the network, store credentials or sensitive data, execute remote code, or bypass OS security boundaries. That is a reassuring operating model for local automation, although granting accessibility control still deserves the same care as any powerful desktop utility.

Pricing

Open source — free (MIT licensed, runs locally, zero per-action cost). There are no paid seats, usage quotas or renewal charges in this plan. It is the right choice for individuals, teams and local-model projects that can manage installation and permissions themselves. The trade-off is not a cheaper tier with fewer features; rather, users take on operational responsibility, and the project’s unmaintained status makes long-term support uncertain.

Platforms

Precompiled binaries cover macOS Apple Silicon and Intel, Linux and Windows, with installation instructions for npm, pnpm, yarn and bun. Desktop control spans macOS, Linux, Windows and Electron applications, and browser control is included. That cross-platform reach is a practical advantage for mixed-device teams, provided each machine grants the required accessibility permissions.

Who it's for

Choose agent-cu for a local command-line foundation behind Claude Code, Cursor, Codex or another supported agent, or for deterministic desktop sequences represented as batched commands and YAML workflows. It also suits humans who prefer terminal-driven control of desktop apps. It is a poor fit for anyone requiring visual screen interpretation, remote deployment, or an actively maintained project with a dependable path for fixes and compatibility updates.

Pros and cons

Pros

• Accessibility-tree references make interface targeting more structured than coordinate-only automation.
• Snapshots, selectors, JSON output, batching and YAML workflows cover both interactive use and repeatable automation.
• Electron CDP support reaches widely used apps including Slack, VS Code and Notion.
• Local execution, MIT licensing and zero per-action cost avoid service fees and network transmission.
• macOS, Linux and Windows binaries reduce installation friction across desktop environments.

Cons

• The repository says it is no longer maintained, which raises compatibility and security-support concerns.
• It requires broad OS accessibility permission to read and control applications.
• It has no screen understanding, limiting tasks that depend on visual interpretation rather than accessible UI elements.
• As a CLI, it expects users or connected agents to handle setup, permissions and workflow design.

Alternatives

For a maintained agent workflow centered on a paid service and broad web, desktop and mobile availability, consider Claude Code. If you want another free, open-source computer-use option that can run through APIs or locally hosted models on Windows and macOS, Computer Use OOTB is a better direction. Gemini fits readers seeking a freemium, cross-platform assistant with features such as Deep Research and Gemini Live.

QuickDesk is a free MIT-licensed alternative with no feature restrictions or connection limits. Screenbox offers free AGPL-3.0 self-hosting. Clawd Cursor is another free, MIT open-source option for Linux, macOS, Windows and self-hosting. Lapu AI provides a free core tier with a paid plan for more usage and workflow scheduling. open-computer-use is a free MIT-licensed local option for macOS, Windows and Linux.

For a wider comparison, browse the AI Computer Use Agents list.

Verdict

agent-cu is best for technically capable users who need free, local, terminal-based desktop control and can connect it to an agent or maintain their own workflows. Its accessibility-tree targeting, Electron coverage and cross-platform binaries are the main reasons to choose it. Look elsewhere if ongoing maintenance, visual screen understanding or a supported commercial service matters more than zero-cost local control.