naïve
← Blog
LaunchOctober 6, 2026·Updated November 5, 2026·7 min read

Introducing the Mobile Sandbox: A Disposable Cloud Phone Your Agents Can Drive

Naive Managed Agents runs disposable cloud Android phones and iPhones that your code and your agents drive by looking and touching. Here is what one is, how it ends itself, and what it costs.

Launch/mobile

TL;DR

  • A mobile sandbox is a disposable cloud phone: an Android 13, 14 or 15 phone ready in about a minute, or an iPhone usually ready in about a minute, up to three.
  • You drive it the way a person does: take a screenshot, tap by point (or, on Android, read the UI tree and tap by an element's text, id or label), type, press a key, scroll, open a URL or deep link, or install an app while it runs.
  • Each phone runs in its jurisdiction (US, EU or Asia) and its traffic leaves from a residential IP in that country, with GPS, timezone, locale and carrier to match.
  • It is the third member of the computer suite, beside the Linux sandbox and the managed browser. It is live today from the API, SDK, CLI and dashboard, where a person can watch the screen, and agents pick one up with a built-in tool.
  • There is no pause and no creation fee. Android costs $0.05 a minute and an iPhone $0.08, network included, billed in whole minutes on one ledger line, and a phone ends itself when idle or when the balance runs out.

Update, November 5: the mobile sandbox now runs iPhones as well as Android phones, installs apps on a phone that is already running with install_app, and sends each phone's traffic out through a residential IP in its jurisdiction. The rate is now per platform and includes that network. This post is revised to match.

Some work only happens on a phone

A growing share of agent work ends at a screen that is not a browser. A mobile build needs a smoke test before it ships. A supplier's app has no web version, so the only way to check an order is to open the app. A crash report says "Android 13" or "iPhone only" and nobody on the team has one to hand. Each of these needs a phone, and a phone on a desk does not scale, does not sit behind an API, and cannot be thrown away when the job is done.

The mobile sandbox is a disposable cloud phone that Naive Managed Agents boots on request and your code drives the way a person does, by looking at the screen and touching it. It is the third member of the computer suite, beside the Linux sandbox and the managed browser. The Mobile sandbox primitive page summarizes the surface; the mobile reference has every field.

One object

A mobile sandbox is a single resource, a mobile device on the wire (/v1/mobile_devices, ids prefixed mob_). It needs no Linux sandbox and has no shell. It is a screen, touch input and an app lifecycle. Android and iPhone are the same object; platform says which.

AndroidiPhone
platformandroid (the default)ios
VersionsAndroid 13, 14 or 15None to choose
Ready inAbout a minuteAbout a minute, up to three
AppsPackage names, at create or with install_appApp Store bundle ids, at create or with install_app
Price$0.05 per device-minute$0.08 per device-minute

Either way a fresh phone is cheap and quick enough that an agent or a test run can create one per task rather than book a shared machine and clean up after it.

Where the phone is

jurisdiction (us, eu or as, us by default) picks where a phone runs, and now also where its traffic comes from. Every phone's traffic leaves from a residential IP in that jurisdiction's country: the United States, Germany or Singapore. Its GPS, timezone, locale and carrier are set to match the exit IP, so an app that checks where it is sees one consistent answer instead of a phone in Frankfurt on a datacenter address.

Look, then touch

Every interaction is one of eight actions sent to the device's act route. They are the same on both platforms (except that most iPhones return no element_tree, so you tap by point, and have no Back or Recents key), and the same whether a script, a person in the dashboard, or an agent is doing the driving.

ActionWhat it does
screenshotReturns the screen as an image
element_treeReturns the UI hierarchy, so a tap can name an element instead of a point (Android)
tapTaps a point, or on Android an element by its text, id or accessibility label
typeTypes text into the focused field
press_keyPresses a key such as Home or Enter (Back on Android)
scrollScrolls the screen
open_urlOpens a URL or a deep link
install_appInstalls an app from the hosted app store on the running phone

Screenshot pixels map one to one onto tap coordinates, so "look at the picture, tap where the button is" works without any scaling math. On Android, when a point is too brittle, tap by selector instead: { "text": "Sign in" } finds the element the way a person reads the screen. Take the next screenshot to see what an action did.

install_app takes the app's id, { "type": "install_app", "app": "com.spotify.music" }: a package name on Android, an App Store bundle id such as com.burbn.instagram on an iPhone. It installs straight from a hosted app store, with no file to upload and no Google or Apple account to sign in, and answers installed, failed with a reason, or installing for a large app still on its way. An agent that reaches a step needing an app the phone does not have installs it and carries on, instead of getting stuck.

Watch it live

In the dashboard, under Computers → Mobile, every device has a workspace: the screen, which you click to tap; the hardware keys (Home and Enter, plus Back and Recents on Android); controls to type, scroll and open a URL; and a Live toggle that swaps the screenshot for a realtime stream of the screen, in the same place and at the same size. That makes a mobile sandbox as useful for a person reviewing an agent's work as for the agent doing it. You can watch a test flow run, step in to dismiss an unexpected dialog, and hand control back.

A mobile device's page in the dashboard: an Android phone streaming live in the screen panel, with the controls and the device's timers, rate and cost so far beside it
An Android device streaming live. The controls beside it drive the same phone.

Timers, not pause

A mobile device has no pause. It lives until you delete it or one of two timers fires, both set at create:

TimerDefaultRange
idle_timeout_minutes: nobody acts on it for this long10 minutes1 to 180
max_duration_minutes: a hard ceiling on its life120 minutes (60 for a phone an agent opens)1 to 1440

Two failsafes sit on top of the timers. Creating a device needs a funded balance, and a device is ended automatically if the organization's balance runs out or its plan lapses. There is no fixed number of phones per account: phones run as long as the account has credit to pay for them. A phone someone forgot about on Friday is not still running on Monday.

Driving one from the SDK and the CLI

The SDK is one method per route.

An iPhone in Germany: install, look, tap, clean up
const { mobileDevices } = naive;
let phone = await mobileDevices.create({ name: "iphone", platform: "ios", jurisdiction: "eu" });
while (phone.status === "creating") {
  await new Promise((r) => setTimeout(r, 5000)); // usually about a minute to boot
  phone = await mobileDevices.get(phone.id);
}
 
await mobileDevices.act(phone.id, { type: "install_app", app: "com.burbn.instagram" });
 
// look.image is the screen as a base64 image
const look = await mobileDevices.act(phone.id, { type: "screenshot" });
await mobileDevices.act(phone.id, { type: "tap", x: 196, y: 400 });
 
await mobileDevices.delete(phone.id);

For an Android phone, pass os_version instead of platform: { name: "pixel", os_version: "15" }. The CLI covers the same lifecycle, which makes it easy to script a smoke test in CI.

The same lifecycle from the CLI
naive mobile create --name iphone --platform ios --jurisdiction eu
naive mobile install mob_… com.burbn.instagram
naive mobile screenshot mob_… --out shot.jpg
naive mobile act mob_… --type tap --x 196 --y 400
naive mobile delete mob_…

The REST routes are the same shape, and they are also published in the platform's MCP catalogue. Details: API, SDK, CLI.

What it costs

There is no creation fee. An Android phone costs $0.05 per device-minute ($3.00 per device-hour) and an iPhone $0.08 per device-minute ($4.80 per device-hour), and both include the phone's residential network: $0.02 of each minute is the network, the rest is the device. A phone is billed in whole minutes from create to end, with a one-minute minimum. It appears as a single mobile line item on the organization ledger, drawn from the same prepaid balance as everything else. Phone minutes bill that balance directly: they are not held against a session's or an agent's budget cap. Because a device ends itself when idle, the bill tracks the minutes a phone is actually in use, not the minutes someone forgot to clean up.

What it is, and what it is not

Being honest about the edges:

  • Agents pick up phones themselves. Beside the API, SDK, CLI and dashboard, an agent on the default harness has a built-in mobile tool. It opens a phone on first use, sees the screenshots it takes, saves them as files, and installs an app when the task needs one. It asks before its first call, and you can watch the phone live while it works.
  • A phone is not bound by a browser's allowed domains. open_url opens any URL or deep link, so keep the tool on approval for an agent whose web reach you have narrowed.
  • Apps come from the hosted app store, by id. Install at create or on a running phone. There is no upload route, so you cannot bring your own build yet.
  • Phones only. Android phones and iPhones, in the cloud. They cover builds, flows and version-specific bugs, not a particular handset's hardware. No tablets or watches yet.

Get started

Create an Android phone or an iPhone from the dashboard's Computers → Mobile page, or with naive mobile create. The full surface is in the mobile reference, and the computer launch post covers the Linux sandbox and browser that a mobile sandbox now sits beside.

Three ways to get started with Managed Agents

Hand it to your agent01

Paste one prompt into Claude Code, Codex, or Cursor and it sets Managed Agents up in your project.

Install the CLI02

Create agents, run sessions, stream events, and see cost from the terminal.

Open the Studio03

Agents, sessions, budgets, and the ledger, all in the browser.

app.usenaive.ai→

FAQ

What is a mobile sandbox in Naive Managed Agents?
A disposable cloud phone, Android or iPhone, that Managed Agents boots on request and your code drives through screenshots, a UI tree and touch actions. It needs no Linux sandbox and has no shell; it is a screen, touch input and an app lifecycle.
Can I get an iPhone?
Yes. Create the device with platform set to ios. An iPhone has no OS version to choose and is usually ready in about a minute, up to three. The live view, the timers and most actions work as on Android, but most iPhones have no element tree, so you tap by point from a screenshot, and there is no Back or Recents key.
Are these real physical devices?
They are phones in the cloud, not a handset on your desk. That covers testing a build, driving an app with no web version, and reproducing a bug on a specific Android version, but not hardware-specific behavior like a particular handset's camera or radio.
How do I get my app onto the device?
Name it by its app id: an Android package name such as com.spotify.music, or an iOS App Store bundle id such as com.burbn.instagram. List ids in apps at create, or send an install_app action to a phone that is already running. Apps come from a hosted app store, so no Google or Apple account is needed. There is no upload route, so you cannot bring your own build yet.
Where does a phone's traffic come from?
From a residential IP in its jurisdiction's country: the United States for us, Germany for eu, and Singapore for as. The phone's GPS, timezone, locale and carrier are set to match that IP, so an app sees one consistent location.
Can an agent use a phone inside a session today?
Yes, on the default harness. An agent there has a built-in mobile tool: it opens a phone on first use, sees the screenshots it takes, installs an app when a task needs one it does not have, and asks for approval before its first call. You can watch the phone live while it works.
What happens if I forget to delete a device?
It ends itself. Every device has an idle timeout (10 minutes by default) and a hard maximum lifetime (120 minutes by default, 60 for a phone an agent opens), and it is also ended if the organization's balance runs out or its plan lapses.