Use case · Games
Let an AI agent play-test your mobile game on real phones
Describe what a player should do — finish the tutorial, clear level 3, open the shop — and Green Phone Agent plays it on real devices, recording video and performance as it goes.
Why game QA is hard to automate
No stable selectors
Game engines render to a canvas, so classic UI automation tools can't see buttons or sprites.
Long, repetitive sessions
Re-playing the same levels after every build eats hours of tester time.
Performance varies by device
Frame drops, overheating and battery drain only show up on real GPUs.
What the agent checks
Tutorials & onboarding
Follows hints, skips intros and verifies each step unlocks the next.
Levels & gameplay loops
Swipes, taps, drags and long-presses through levels and checks win/lose states.
Rewards & pop-ups
Daily rewards, level-complete dialogs, offers and in-game shop screens.
Performance
Average FPS, heat and battery use per session on real hardware.
UI on many screens
Notches, aspect ratios and small screens that cut off HUD elements.
Crashes & freezes
Every crash comes with the video timestamp and steps to reproduce.
Example scenarios
Written in plain language — exactly what you'd tell a human tester.
- 1Install the build, complete the tutorial and check the player reaches the main map.
- 2Play level 1 to 3, collect all coins and verify the coin counter matches the reward screen.
- 3Open the shop, browse every tab and check no item image is missing.
- 4Play for 15 minutes and report average FPS and device temperature.
Frequently asked questions
Do we need to add an SDK or hooks to our game?
No. The agent looks at the rendered screen and acts like a player, so it works with Unity, Cocos, Unreal and other engines without code changes.
Can it handle fast real-time gameplay?
It's best at turn-based, puzzle, casual and menu-heavy flows. For fast action games it's great for onboarding, menus, rewards and performance runs.
Which devices can we test on?
Physical Android and iOS phones in OneGreen's lab, including mid-range models where performance problems usually show up first.
Use cases
Let an AI agent run your next test cycle
Book a 30-minute demo. Bring a build and a few test scenarios — we'll run them live on real phones.