OS automation for AI agents.

Drives desktop apps through their accessibility tree. Compact text output, ref-based, 100% native Rust.

$npm install -g @agent-ctrl/cli

Agent-first.

Compact text output uses fewer tokens than JSON or DOM. Refs let agents target elements deterministically without re-querying.

Accessibility-first.

Snapshots the a11y tree, not pixels. Stable across themes, locales, and resolutions where vision models drift.

Cross-platform.

UIA on Windows, AX on macOS, AT-SPI on Linux. One schema, native binaries, no runtime.

Fast.

Native Rust binary. Instant command parsing, zero scripting overhead. No drivers, no Node, no browser.

Composable.

Same schema whether you are scripting, inspecting, or feeding an LLM. Pipe snapshots and refs through any tool that reads stdin.

View on GitHub

agent.tsrun with agent-ctrl v0.1.1

1import { AgentCtrl } from "@agent-ctrl/client"
2 
3const ctrl = new AgentCtrl()
4const session = await ctrl.openSession("uia")
5 
6// snapshot the current a11y tree
7const snap = await ctrl.snapshot(session)
8console.log(snap.refs.entries)