Agent Device

agent-device gives AI coding agents a live feedback loop on a running app, so an agent can check its own changes. Agents inspect, control, debug, and verify apps on iOS, Android, and HarmonyOS (simulators, emulators, and physical devices), plus tvOS, Android TV, Amazon Vega OS TV, web, macOS, and Linux. It is exposed three ways over one runtime: a CLI, a built-in stdio MCP server, and a typed Node.js API.

Instead of reasoning over screenshots alone, agents read token-efficient snapshots of the app's accessibility tree, where each element gets a stable ref such as @e2. They act through those refs or through semantic selectors by text, label, or role, and commands that wait for the UI to settle return a diff of what changed. Screenshots, video, logs, traces, network data, performance samples, crash details, and React profiles serve as evidence for review.

Device state lives in sessions scoped to the caller's git worktree, and host-local device claims stop parallel agents from taking over each other's simulators. Each platform uses its own backend: XCTest and an accessibility bridge on iOS and tvOS, ADB on Android, HDC and ArkUI on HarmonyOS, AT-SPI on Linux, and Agent-browser for basic web support. Newer backends cover a subset of commands. The same commands drive hosted devices on BrowserStack, AWS Device Farm, and Limrun.

Working runs can be saved as replay scripts for local use or CI, with experimental healing of stale selectors, or exported as Maestro YAML. Unlike Appium, Detox, or Maestro, where teams write and maintain test suites, the agent chooses each command at run time. It supports native iOS and Android apps as well as React Native, Expo, and Flutter apps.