At a glance
Everything about Agent Device is public. The README and the docs at oss.callstack.com cover installation, commands, selectors, replay, device clouds, and known limitations. The license in the repository is MIT.
- Approach
- AI agent with vision, no selectors
- Writing a test
- Plain-English steps, CLI, MCP, or dashboard
- Native mobile
- iOS simulators + Android emulators
- Public pricing
- Free tier, then $99/mo self-serve
- Pull requests
- GitHub App: run per PR, check + comment
- Approach
- Your coding agent picks each command, no AI inside
- Writing a test
- CLI, MCP server, or Node.js API, plus .ad replay scripts
- Native mobile
- iOS, Android, HarmonyOS, TV, and physical devices
- Public pricing
- Free under MIT, with your own devices and agent tokens
- Pull requests
- Your CI replays saved scripts, Expo EAS template included
The same test, both ways
Eleven Agent Device commands sign up a new iOS user, with the email code pasted in from outside. Five TesterArmy sentences do the same. The agent reads the code itself.
1+added: # workflows/signup.sh - Agent Device on the iOS simulator2+added: agent-device open com.acme.shop --platform ios3+added: agent-device find text "Create account" click4+added: agent-device find label "Email" fill "ada@example.com"5+added: agent-device find label "Password" fill "Str0ng-Passw0rd!"6+added: agent-device find text "Continue" click7+added: agent-device wait text "Enter the 6-digit code" 50008+added: # The code arrives by email. Paste it in as OTP_CODE.9+added: agent-device find label "Code" fill "$OTP_CODE"10+added: agent-device find text "Verify" click11+added: agent-device wait text "Welcome, Ada" 500012+added: agent-device screenshot ./evidence/signup-done.png13+added: agent-device close
Where they differ
Agent Device drives a device you provide and does what your coding agent tells it. TesterArmy provides the agent, the devices, and the verdict as one service.
A tool or a platform
Agent Device is a CLI your coding agent runs against your own device. TesterArmy is a hosted service with its own agent, devices, and verdicts.
Who runs the device
Agent Device drives simulators on your Mac or CI runner, so you install Xcode and the Android SDK. TesterArmy runs cloud simulators and emulators.
Codes from email and SMS
TesterArmy gives each project an inbox and a phone number, so the agent reads verification codes itself. With Agent Device, you supply the code.
How a run repeats
Agent Device saves a session as a .ad script that replays without AI. TesterArmy re-reads the plain-English steps each run and looks at the screen.
Side by side
Agent Device gives a coding agent a token-efficient way to see and operate a real app, physical devices included, for free. The rows below set that against a hosted platform.
- How you startTesterArmySign up, upload a simulator build or enter a URL, write the first flow.Agent Devicenpm install, doctor check, then hand the CLI or MCP server to your agent.
- Who decides each stepTesterArmyTesterArmy's own agent reads the plain-English step and drives the screen.Agent DeviceYour coding agent picks every command. Agent Device has no AI inside.
- Where runs happenTesterArmyCloud iOS simulators and Android emulators, plus a hosted browser for web.AdvantageAgent DeviceYour Mac or CI runner, with Xcode or the Android SDK installed.
- Physical devices and TVTesterArmySimulators and emulators only, with no physical devices and no TV.Agent DevicePhysical iOS and Android devices, tvOS, Android TV, HarmonyOS, macOS, Linux.Advantage
- Email and SMS codesTesterArmyEach project gets an inbox and a phone number the agent reads itself.AdvantageAgent DeviceThe CLI types the code. Your agent must fetch it from the email itself.
- Pull request verdictsTesterArmyA GitHub App check and comment with the video, plus an exploration agent.AdvantageAgent DeviceYou wire replay scripts into your CI. Callstack publishes an Expo EAS template.
- Evidence and debuggingTesterArmyStep trace, screenshots, and an MP4 synced to every step.Agent DeviceScreenshots, video, logs, network, traces, CPU and memory, React profiles.Advantage
- PricingTesterArmyFree tier to start. Hobby is $99 a month with 250 runs.Agent DeviceFree under MIT. You pay for devices, clouds, and your agent's tokens.
Facts checked September 2026. Spotted something out of date? Tell us and it gets fixed.
Pricing
Agent Device is free software. You pay for the Mac, CI minutes, or device cloud it drives, plus your agent's tokens. TesterArmy bills $99 a month beyond the free tier.
$99/mo after a free tier
- Free tier: sign up and run against your own app
- Hobby $99/mo: up to 250 test runs included, web and mobile
- Startup $299/mo: up to 1,000 test runs included, 10 concurrent test runs
Free and open source under MIT
- The CLI, MCP server, and Node.js API are MIT licensed on npm and GitHub
- Local runs need a Mac with Xcode, or the Android SDK on Mac or Linux
- Remote devices come from BrowserStack, AWS Device Farm, or Limrun on their own plans
Who should pick which
Many teams run both. Agent Device checks a screen while the coding agent builds it. TesterArmy checks the pull request and runs the regression suite.
- Pick TesterArmy ifYou want tests your product managers and QA engineers can read and edit without a terminal.Pick Agent Device ifYour coding agent should verify its own change on the simulator before it opens a pull request.
- Pick TesterArmy ifSign-up, login, and checkout flows depend on email or SMS codes the agent must read itself.Pick Agent Device ifYou test on physical devices, tvOS, Android TV, HarmonyOS, macOS, or Linux desktop apps.
- Pick TesterArmy ifYou want a pass or fail check on every pull request without building the pipeline yourself.Pick Agent Device ifYou want an MIT-licensed tool inside your own CI and device cloud, and no vendor account.
Questions
Run the same flow on a cloud device
Take the screen your coding agent verified with Agent Device this week, describe that flow in sentences on the free tier, then let the pull request check run it.