Argent gives a coding agent direct control of running applications on devices: iOS simulators and physical iPhones, Android emulators and devices over the Android debug bridge, television platforms including tvOS, Android TV and Fire TV, and Electron or Chromium desktop and web applications over the Chrome DevTools Protocol. Each target gets the interaction model that suits it — touch and gestures on mobile, a directional remote on television, mouse and keyboard on desktop.
With that access an agent can build, launch and drive an application the way a person would, then read back what happened: tapping, swiping, pinching, typing, gestures and hardware buttons, navigation from accessibility-tree coordinates, deep links, and multi-step sequences issued in a single call with feedback after each interaction. A sequence can be captured once and replayed deterministically, turning a manual reproduction into a repeatable check, and screenshots can be diffed against a saved baseline with comparison that accounts for text and fonts.
Debugging and profiling happen in the same session. The agent can attach a debugger, read console logs, evaluate expressions in the running application, walk the native and React component trees, and capture network traffic at both the JavaScript request layer and the native layer. Profiling covers React Native through Hermes, React DevTools, and native tooling — Xcode Instruments and Android Perfetto — down to fibre renders, CPU hotspots and correlated commit-versus-hang reports, and one recorded session can be queried repeatedly without re-running it.
| Platform | Targets | Interaction |
|---|---|---|
| iOS | Simulators and physical iPhones | Touch and gestures |
| Android | Emulators and physical devices over adb | Touch and gestures |
| TV | Apple TV, Android TV and Google TV, Amazon Fire TV | Directional remote |
| Desktop and web | Electron and Chromium apps, including React Native Web and Expo web | Mouse and keyboard |
Setup registers an MCP server with whichever editor is detected and copies skills, rules and agent definitions into the workspace; the configuration can be global to the machine or committed with the project so a team shares one setup. Editors covered include Claude Code, Cursor, VS Code, Windsurf, Zed, Gemini CLI, Codex CLI, Hermes, opencode and Kiro. React Native projects need no extra wiring, and on Linux the tool selects emulator graphics and virtualisation settings that avoid silently corrupted emulator output.
Features
- Multi-platform control: iOS, Android, three television platforms and Electron or Chromium applications through one toolkit
- UI interaction: taps, swipes, pinches, typing, gestures, hardware buttons, remote navigation, mouse and scroll
- Record and replay: captured interaction sequences replayed deterministically for reproductions and smoke checks
- Visual regression: screenshot diffing against a baseline, with text- and font-aware comparison
- Profiling: React Native, React DevTools and native Instruments or Perfetto sessions, re-queryable after recording
- Runtime inspection: console logs, expression evaluation, native and React component trees, and network traffic at both layers
- Desktop and web control: tabs, cookies, storage, the document tree and network over the Chrome DevTools Protocol
- Editor registration: MCP configuration written for the detected editor, globally or as a committed project dependency
- React Native support: build, launch and iterate on a React Native project without additional setup
- Linux emulator defaults: graphics mode and virtualisation checks chosen to avoid slow or blank emulator sessions
Integrates with
Claude Code
Terminal-based agentic coding tool that reads your codebase and runs tasks through natural-language commands
Codex
Lightweight coding agent that runs in your terminal
Hermes
Self-improving AI agent with a learning loop that creates and refines skills, recalls past sessions, and runs across chat platforms
OpenCode
Open-source AI coding agent that runs in the terminal, an IDE, or a desktop app
