Perception → Decision → Action
The noqa agent operates entirely from screenshots — it doesn’t inspect view hierarchies, source code, or accessibility trees. On every step:- Perceive — capture a screenshot and analyze what’s visible: buttons, text, images, current state.
- Decide — determine the next action to take toward the test goal.
- Act — execute the action and observe the result.
Available actions
| Action | Description |
|---|---|
| Tap | Tap any visible element — buttons, links, list items; supports double tap and long tap |
| Swipe | Swipe in any direction to navigate or dismiss dialogs |
| Drag | Drag an element from one position to another |
| Input text | Clear a field and type the specified text |
| Open link / deeplink | Navigate directly to a URL or in-app deeplink |
| Terminate app | Force-stop the app |
| Background app | Send the app to the background |
| Activate app | Bring the app back to the foreground |
Understanding instructions
The agent understands natural language — you don’t need to describe every tap and swipe. Write at the goal level and the agent figures out the individual steps. For example, if your app has a long onboarding with 20–30 screens but each one has just one or two buttons, you can simply write “Complete onboarding” — the agent will go through all the screens on its own without any further instructions.Speed
On the first run, the agent takes 5–10 seconds per action — it needs to analyze the screen and reason about the next step. On subsequent runs with similar screens, this drops to 1–2 seconds when the agent has already seen similar screens. In practice, the agent executes most test flows faster than a manual tester would.Limitations
- Timing — the agent takes a screenshot, analyzes it, then acts. If something happens fast — a toast, a transition, a brief loading state — the agent may miss it or not react in time.
- Test length — we recommend keeping tests under 100 actions. Longer tests increase the chance of the agent losing context and making incorrect decisions.
- Device type — available testing capabilities depend on whether you’re using a local device, simulator, or cloud device. See Local vs Cloud for details.
- Platform — cloud tests can be started from the web dashboard. Local tests require a Mac with the noqa desktop app installed.