Skip to main content

Watch a coding agent use BugRadar sandboxes with the Fitney app.

A sandbox is a cloud iOS Simulator, Android Emulator or real device that your coding agent drives with the BugRadar CLI. Watch sandboxes live, or replay them, on the Sandboxes page.

Start with your coding agent

Paste this into Claude Code, Codex, Cursor or another coding agent:
The agent learns the commands from BugRadar sandbox --help.

Set up

Install the CLI (or run BugRadar update), then log in from your project’s folder:
The login is saved per git repository. BugRadar whoami shows which workspace is in use.

Choose the app

BugRadar sandbox start leases a device, then installs and launches your app. Pick the app with: --device "iPhone 17 Pro" --os 26.0 picks the model and OS (default: latest).

Builds

See Building your app. After a native change, run BugRadar sandbox upload <path>. --reset installs clean.

Commands

Commands after start act on the current sandbox, one per git worktree. --sandbox <id> picks another. Once a worktree has more than one sandbox, every command needs --sandbox <id>, so agents sharing a worktree never drive each other’s device.
  • Each action waits for the screen to settle, then saves a screenshot to .BugRadar/.sandboxes/<sandbox-id>/.
  • Coordinates are pixels in the latest screenshot.
  • --description "Open settings" titles the step on the sandbox’s page.
  • Every command takes --json.

Local development

forward connects ports on the device to the same ports on your machine, so the app reaches your local backend at localhost:<port>:
  • 3000:8000 sends the device’s port 3000 to your port 8000.
  • Metro: start Metro, then forward its port. Saved changes hot reload. Debug builds, including Expo development builds, load from it.
  • Native changes still need BugRadar sandbox upload <path>.

Logs, network, performance and video

Each command takes --since <step> and saves its output next to the screenshots. After a sandbox ends, logs, network and perf with --sandbox <id> show what it kept, step by step. The kept log is the app’s own, so --all doesn’t apply.

Watch and replay

start opens the sandbox’s page in your browser (--no-open skips it). The Sandboxes page lists live and recent sandboxes.
  • Live: each step appears as it happens. Extend keeps the sandbox longer.
  • Ended: replay each step with its screenshot, a screen recording, and Performance, Logs and Network tabs.

End a sandbox

Run BugRadar sandbox end when you’re done.
  • A sandbox ends 30 minutes after its last command, and lasts at most 12 hours.
  • BugRadar sandbox extend 1h holds it through a long pause, like a native build.
  • Your plan limits how many sandboxes can be live at once.

Real devices

start --physical uses a real iPhone or Android device. Message us on Slack or email us to turn this on. On a real device:
  • There’s no forward, so point the app at public URLs.
  • upload always installs clean.
  • On a real iPhone, performance and the screen recording are only available after the sandbox ends. BugRadar sandbox perf --sandbox <id> shows performance, and BugRadar sandbox video --sandbox <id> saves the recording. A real device’s recording arrives a few minutes after the end.
  • Idle and total time limits are shorter. BugRadar sandbox status shows the time left.

Add it to your agent instructions

To have your agent check its work in a sandbox every time, add this to its instructions file (AGENTS.md, CLAUDE.md, etc.):