Skip to main content
noqa works with any game engine — Unity, Unreal, Godot, or anything else. Since the agent operates entirely from screenshots, it doesn’t care what’s rendering the UI. There’s no instrumentation, no SDK, no framework-specific setup required.

​
What works well

The agent is a good fit for testing games where actions happen at a pace it can keep up with — roughly one interaction every 5–10 seconds. This covers most menu flows, onboarding, in-app purchases, progression screens, and turn-based mechanics.

​
What doesn’t work

Real-time games that require fast, continuous input — like action shooters or rhythm games — are not a good fit. The agent takes a screenshot, thinks, then acts. It can’t react fast enough to keep up with real-time gameplay. Complex game logic is also a challenge. If the agent needs to understand rules, make strategic decisions, or figure out what to do next without any visual cues, it may struggle or make wrong choices.

​
How to make testing easier

The most effective approach is to adapt the game slightly for testing — not to change the product, but to give the agent clear signals about what to do next. For example: show a visual hint or highlight the next expected tap target. The agent doesn’t need to reason about the game state — it just needs to see where to tap. This dramatically improves reliability and reduces the number of failed runs caused by the agent getting confused rather than the game actually breaking. Think of it as reducing the cognitive load on the agent: the less it has to figure out on its own, the more reliably it executes the test.

​
Speed

In regular apps, the agent can speed up on repeated runs because it recognizes screens it has seen before — dropping from 5–10 seconds per action to 1–2 seconds. In games, this typically doesn’t happen: each screen is unique and dynamic, so the agent has to analyze every frame from scratch. Expect 5–10 seconds per action consistently when testing games.