agent-device is a device automation CLI for AI agents. It is built for real apps on iOS, Android, TV, and desktop, and its stated aim is mobile verification for AI agents. The homepage frames it around giving agents a structured way to understand and act on native app screens.
A central feature is snapshot-based interaction. The snapshot command exposes the accessibility tree and assigns stable references such as @e2, with output described as token-efficient and smaller than raw screenshots or verbose dumps. The tool also supports semantic selectors, allowing elements to be found by text, label, or role. Example commands shown on the page include opening apps, taking snapshots, pressing referenced elements, filling inputs, replaying saved sessions, and using an experimental -u flag to heal stale selectors during replay.
agent-device is presented as usable across development simulators, physical QA devices, and TV-class targets. The page shows example use on iOS and Android, including an emulator serial and a TV target, and says the same workflow moves across iOS, Android, and TV. It also mentions running with a local agent or from a cloud sandbox, and provides an agent-device proxy for remote access. Beyond the CLI, there is a fully typed TypeScript client for direct integration into codebases or custom cloud agents, with examples for managing sessions, provisioning simulators, opening apps, and capturing snapshots programmatically.
The terminal is also used for debugging and evidence capture. The page lists parsed logs, network dumps, screenshots, and screen recordings, plus commands for changing device settings and sending push notifications. It also describes official skills packages for canonical workflows, targeting rules, debugging patterns, and replay guidance. Installation is shown as an npm global install with npm install -g agent-device, and the page says it is made by Callstack, identified there as React and React Native development experts.
agent-device is a CLI tools & terminal project. It focuses on automating and verifying real device interactions for AI agents across mobile and desktop platforms is complex and manual. It is built as an open-source project for AI agent developers and automation engineers. It ships for the command line.
agent-device first shipped in 2026. The project is developed in the open on GitHub with 3k stars and 510 commits in the last 90 days. Key capabilities include device automation, snapshot capture, and semantic selectors. It exposes integrations via a public API.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do