Skip to main content
This guide walks you through every step of running a test on Duzz — from initial project setup to reviewing results. Follow these steps in order and you’ll have your first test running in minutes.

​
Step 1: Create a Project

A project is the top-level container for everything — builds, devices, scenarios, and runs.
Every test starts with a project. A project typically maps to one game or app.
  • Name — A clear name for your game or app (e.g., “QuickHits Casino”, “Animals & Coins”)
  • Prefix — A short 2–5 letter code (e.g., QHC, INNP). Shows up on run IDs to keep things organized. If you skip it, one is generated from the project name
  • Description — Optional notes about the project
Think of the project as the home base for everything related to one game. All builds, scenarios, devices, and results live under it.

​
Step 2: Upload a Build

A build is a specific version of your game that you want to test.
Before you can run any test, you need at least one build uploaded.
  • Label — A human-readable version name (e.g., “v2.4.1-staging”, “Release 3.65”)
  • Platform — Android, iOS, or Web
  • File — The app binary (.apk for Android, .ipa for iOS, or a URL for web)
  • Build number — Optional numeric identifier from your CI system
After uploading, the build goes through a short processing step. Once the status shows Ready, it can be used in runs.
Build StatusMeaning
ProcessingBuild is being validated and prepared
ReadyBuild is available for runs
ErrorSomething went wrong during processing — check the file format and re-upload
Make sure the build platform matches the device you plan to run on. An Android build won’t work on an iOS device.

​
Step 3: Set Up Devices

Devices are the real phones or emulators your tests will run on.
Duzz uses a cloud device farm to provide remote devices. You allocate devices to your project from the available inventory.
  • Device name — e.g., “Samsung Galaxy S23”, “iPhone 15”
  • Platform — Android, iOS, or Web (must match your build)
  • Manufacturer / Model / OS — Helps you pick the right device for your testing needs
  • Availability — Shows whether the device is currently available in the farm
You don’t need to configure the device itself — Duzz handles installing the build, launching the app, and setting up the video stream automatically.
Allocate at least one device per platform you want to test on. You can always add more later for parallel testing.

​
Step 4: Configure the Agent

The agent template defines how the AI agent behaves during test runs.
Every project has at least one agent template. This is where you shape the agent’s personality and knowledge.
  • Name — A label for this agent configuration (e.g., “Default Agent”, “Casino Expert”)
  • Guidelines — Behavioral rules for the agent:
    • “Always dismiss cookie banners before testing”
    • “Accept terms of service when prompted”
    • “Decline or skip all login/sign-up requests”
    • “Close pop-ups unless instructed to interact with them”
  • Capabilities — The tools the agent can use during every scenario (tap, scroll, swipe, type text, etc.). These are set once at the agent template level — you can’t choose different tools per scenario or per run
Most users start with the default agent template and only customize guidelines once they see how the agent behaves with their specific game.
Guidelines are your most powerful lever. If the agent keeps doing something you don’t want (like tapping on ads), add a guideline to correct it.

​
Step 5: Write Your Knowledge

Knowledge is the game context document that tells the agent what your game is and how it works.
This is a markdown document that describes your game at a high level — mechanics, UI layout, terminology, expected behaviors. It’s always included in every run.
  • Describe each screen the agent might encounter
  • Explain key game mechanics and terminology
  • Note any non-obvious UI patterns (hidden menus, long-press actions, swipe gestures)
Knowledge is versioned and tied to builds, so you can track how your game documentation evolves alongside your app.
Good knowledge dramatically improves agent performance. The more the agent understands about your game, the better it navigates and tests.

​
Step 6: Create Skills

Skills are short behavioral instructions that tell the agent how to handle specific situations.
While knowledge describes the game world, skills describe specific actions or techniques.
  • Name + Description — e.g., “Cheat Menu Usage”, “Video Ad Handling”
  • Body — Step-by-step instructions for a specific behavior
  • Scope — Can be set as project defaults (always included) or selected per scenario
Examples:
  • “How to use the debug tool: tap Settings, scroll down, tap Debug, enter the code…”
  • “When a video ad appears, wait for 15 seconds, then look for the X button to appear and close it”
Skills are optional but powerful. If the agent keeps struggling with a specific game mechanic, write a skill for it.

​
Step 7: Create a Scenario with Checkpoints

A scenario is the actual test plan. Checkpoints are the milestones the agent must reach.
This is the core of your test definition. For each scenario:
  1. Give it a clear name — e.g., “Complete First Island”, “Purchase Shop Item”
  2. Set preconditions — What state should the game be in when the test starts?
  3. Add checkpoints in order — Each checkpoint needs:
    • Goal (required) — What the agent should achieve: “Close the daily bonus popup”
    • Instructions (optional) — How to achieve it, if the path isn’t obvious
    • Timeout — How long the agent should try (default: 5 minutes)
Start simple. Your first scenario should have 2–3 checkpoints testing one straightforward flow. You can always add complexity later.

​
Step 8: Add Verifications (Optional)

Verifications are automated quality checks that run after the test completes.
While checkpoints track whether the agent reached each goal, verifications check whether the game behaved correctly along the way. For each verification, add assertions — natural-language questions evaluated against the run’s recorded artifacts:
  • Query — “Is the HUD fully visible and not overlapping other UI elements?”
  • Scope — Which part of the run to examine (specific checkpoint, time window, or the entire run)
  • Severity — How serious a failure would be (blocker, critical, major, minor, info)
Verifications are optional for your first run. Start with checkpoints only, then add verifications once you’re comfortable with the basics.

​
Step 9: Start the Run

You’re ready. Start a new run by selecting:
  1. Scenario — The test plan you just created
  2. Build — Which version of your game (must be in Ready status)
  3. Device — Which device to run on (platform must match the build)
  4. Skills — Which skills to include (project defaults are pre-selected)
  5. Mode — Formal run or Playground
    • Formal run — Runs to completion, device released automatically. Best for structured testing
    • Playground — Agent pauses when done, device stays alive for you to explore. Best for debugging and iteration
Hit Start and the run begins.

​
Step 10: Monitor and Review

Once the run starts, you can watch it in real time:
  • Live device screen — See exactly what the agent sees, streamed to your browser
  • Interactive control — The device is fully interactive. Click, drag, and swipe on the screen to help the agent if it gets stuck — it will continue on its own from wherever you leave it
  • Checkpoint progress — A live checklist showing which milestones have been reached
  • Agent reasoning — The agent’s thought process in real time
When the run completes, review the results:
  • Pass/Fail status — Did all required checkpoints pass?
  • Video recording — Full replay of the device screen
  • Checkpoint details — Screenshot and timestamp for each checkpoint reached or failed
  • Verification results — If you added verifications, see which assertions passed or failed with supporting evidence
  • Agent logs — The full reasoning trace and action history

​
Quick Checklist

StepWhatWhereRequired?
1Create a projectProjectsYes
2Upload a buildBuildsYes
3Set up devicesDevicesYes
4Configure the agentAdmin → Agent TemplatesYes (defaults usually work)
5Write knowledgeKnowledgeRecommended
6Create skillsSkillsOptional
7Create a scenarioScenariosYes
8Add verificationsScenario → VerificationsOptional
9Start the runScenario → New RunYes
10Monitor and reviewRun screen—