Agent Device is a mobile-native automation layer designed for AI agents to interact directly with real iOS and Android applications. It provides a structured interface for exploring, understanding, and controlling mobile UIs through accessibility trees rather than brittle image-based or coordinate-based automation.

The system works by exposing a snapshot of the app’s current UI state as a structured accessibility tree, where interactive elements are assigned stable identifiers. AI agents can then use these identifiers to perform precise actions such as tapping, typing, scrolling, and navigating through the app. This makes mobile automation more reliable, token-efficient, and reproducible compared to traditional testing or screen-scraping approaches.

Agent Device also supports full lifecycle workflows including app launch, session management, replayable execution scripts, and debugging tools like logs, screenshots, and network inspection. These capabilities allow exploratory agent behavior (discovering UI flows) to be converted into deterministic, replayable automation scripts for testing and regression purposes.

It is designed to bridge the gap between AI agents and native mobile environments, enabling use cases such as automated QA, bug reproduction, and end-to-end UI testing driven by LLM-based agents.

Key features include:

• Structured accessibility-tree snapshots for UI understanding
• Deterministic UI actions (tap, type, scroll, find, navigate)
• Cross-platform support for iOS and Android devices/simulators
• Replayable session recording for regression testing
• Debugging tools including logs, screenshots, and network capture

Common use cases include:

• Automated mobile app testing
• AI-driven QA workflows
• Bug reproduction and debugging
• Regression testing via replay scripts
• Enabling AI agents to interact with native mobile apps in a controlled environment

Agent Device is developed by Callstack as an open-source tooling system focused on agent-driven mobile automation.