Expo Simulators give coding agents cloud iPhones and Android phones

Felix Laurent 13 Nov 2026 4 min read

AI agents

Expo Simulators is a new cloud service that gives AI coding agents their own iOS Simulators and Android Emulators. Expo launched it on November 12, and it is now generally available on all paid plans. An agent runs its change on a cloud phone and returns a screen recording before it asks a human to review the work.

Expo is the company behind the popular open-source framework for building mobile apps with React Native. Its pitch is simple: "Coding agents write mobile code faster than any team can review it." Expo Simulators is meant to close that gap.

Why coding agents need cloud simulators

Agents that check their own work give better results, Expo says. On a laptop, an agent can already open a local simulator and test its change. But that setup breaks down once teams run many agents at once.

Expo lists four problems with local simulators:

Limited parallel runs. Seven agents on one laptop need seven simulators, and they eat memory and compute.
Collisions. Agents that share one simulator get in each other's way, for example by reinstalling the app mid-test.
No history. When an agent does something strange, there is no record of what happened on the simulator.
Wrong place. More agents now run in cloud sandboxes and CI, away from the laptop. The iOS Simulator also needs a Mac, which not every developer has.

Expo Simulators moves the phone to the cloud.

How Expo Simulators works

Each session runs on its own virtual machine, so parallel agents never share a simulator. An agent installs a build, drives the app and brings back proof. It uses the same commands it would use on a local simulator, through tools such as agent-device, Argent or Appium.

Any agent that can run command-line tools works with it, including Claude Code, Codex and Cursor. The easiest setup is an Expo skill called eas-simulator. It teaches the agent to check access, start a session, test the app, collect evidence and shut the session down.

Every session records a lot of data:

a video of the full run;
screenshots;
logs and network traffic;
CPU and memory usage.

People can follow along in a browser on desktop or mobile. If something looks wrong, a developer can take over and tap through the app by hand, with no Mac needed.

What changes in code review

The recording is the main point. Instead of reading 500 lines of code that "look about right", a reviewer watches the change work. Nobody has to pull the branch, build the app and open a simulator. The agent can attach the recording and screenshots to the pull request.

Expo gives one example. An agent ran an app at the largest text size, found labels that overflowed and covered each other, and fixed the layout. It then attached before and after screenshots, so the reviewer saw the fix without running a build.

Product managers and designers benefit too. They can open a link to a running session and tap through a new build in their browser, with nothing to install.

Expo also describes how teams already use the service:

Agentic CI, where an agent tests the flows from each pull request and posts a recording;
bug triage, where an agent reproduces a TestFlight bug, fixes it and opens a pull request with before and after shots;
end-to-end tests with e2e, Maestro or Appium suites;
iOS development without a Mac, from Linux, Windows or a cloud sandbox.

The service is not limited to Expo apps. Any app built for the iOS Simulator or Android Emulator runs in a session, including SwiftUI, Jetpack Compose, Flutter and Unity apps.

Security, limits and pricing

Unreleased builds run in the cloud, so Expo stresses isolation. Each session uses the same isolated infrastructure as Expo's Build, Submit and Workflows services, which already handle millions of jobs. No session runs longer than two hours. Teams can set a shorter limit and an idle timeout, so a crashed CI job cannot leave a simulator billing forever.

When a session ends, its simulator disappears. The recording, screenshots and logs stay on the session page for review. Expo Simulators does not offer real physical devices.

Pricing is usage-based: $0.10 per simulator minute for both iOS and Android, or $6 an hour. The Starter, Production and Enterprise plans include 60 free minutes a month until 2027.

Cloud sandboxes for agents are a fast-growing market. Replit, for example, runs agent work in isolated microVMs, and Expo now adds a phone to that kind of setup.

Do coding agents really need cloud phones

For mobile teams that run several agents, yes. Testing is the slowest part of agent-written mobile code, and Expo Simulators attacks exactly that step. Separate VMs remove collisions, recordings make reviews faster, and Windows or Linux users finally get an iOS Simulator without a Mac.

The costs and limits are clear. At $6 an hour, many long agent runs add up, and simulators still miss problems that only appear on real hardware. Teams with one agent on a powerful Mac can stay with local simulators, as Expo itself admits. For everyone else, a recording attached to every pull request is a strong reason to try it.