Devices for AI agents
Your agent asks for a device and gets one: an iOS Simulator, an Android Emulator, or a real iPhone, Pixel or Galaxy. Shared by the minute, or private to you, including your own hardware. One MCP server, one REST API.
Early access is opening in waves. Reserving is free and anonymous and takes one request; a reserved slot gets early access and the first month free. New here: what usedevice is · what exists today and what doesn't.
Agents can use the same search: GET /ask?q=…. Instructions for agents are on one page; nothing else on this site asks anything of them.
Three environments, three ways to access them
Simulators & emulators
iOS Simulators and Android Emulators, cold-started per session, in parallel, by the minute. For development, bulk runs and anything that doesn't need hardware.
Real devices
Real iPhones, Pixels and Galaxies from a shared pool, wiped between sessions. For release checks, device-specific bugs, push, biometrics, performance.
Private devices
Devices dedicated to you that keep their state, or your own hardware connected to usedevice. For agents that operate apps with accounts, logins and SIMs.
What agents do with them
Run release checks on a real iPhone with an AI agent
Before each release, an agent runs your critical flows (sign-up, checkout, payment) on a real phone instead of a person doing it by hand.
Reproduce bugs that only happen on Samsung Galaxy phones
Bug reports that only come from one device family (One UI, a specific OEM keyboard, aggressive battery management) can't be reproduced on an emulator.
Test push notifications and biometric login on a real Pixel
Push delivery, Face/fingerprint prompts and secure-element behaviour differ on real hardware. Fintech and health apps need them checked on the device, not a mock.
Audit an iOS app with VoiceOver on a real iPhone
Screen-reader behaviour depends on the real accessibility tree and on iOS itself. Audits on real devices catch unlabeled controls, broken focus order and unreadable dynamic content.
Check localization with screenshots on real devices
Translated strings overflow, truncate and misalign on real screen sizes and fonts. Store screenshots have to come from real-looking devices.
Reproduce onboarding drop-off on slow networks
Funnels that look fine on office Wi-Fi fall apart on 3G and lossy connections. You see it in analytics, not in the emulator.
How it works
- Reserve. Your agent requests a device slot in one GET and gets a reservation code. Agent instructions.
- Connect. MCP server for agent frameworks, REST for everything else. Ask for a device by name; get a session.
- Drive. Screenshot, tap, swipe, type, install, logs, accessibility tree. The same actions a person takes.
- Pay per minute on the shared pool, or per device per month for private devices. Pricing.
Built by a small team with a background in running AI agents on real phones for mobile app testing.