Getting Started
Create one meaningful check, inspect its evidence, and repeat it on your next release
Start with one flow you repeatedly verify, such as signing in or completing a core task. Your first goal is a reviewed test of the expected behavior and evidence from running it.
1. Prepare the environment and install
Local automation uses the directly distributed macOS app. See the app installation page for available installers and supported macOS versions and hardware. The Mac App Store distribution does not provide local automation.
The installer includes the execution engine and browser. Tests and authentication for your service still need preparation:
- A staging URL and test accounts
- Steps and expected outcomes, such as reaching the correct dashboard after signing in
- A supported model provider account and authentication if using AI
See AI data and costs.
2. Create a project
From the project home, choose New project, enter its name and environment URL, and check the default environment. Name your test environment staging or similar. Open the project and use Overview to continue.
The first-run modal explains the workflow; it does not create the project or run tests for you. Prepare environment values through Settings → Environments, and manage account-specific conditions in Settings → Personas. Creating a project does not sign you into the target service.
Inspect product structure first
Collect pages in Service map → Screens, then organize expected behavior in Features and User flows and link tests. Use discovered pages to choose important user flows; collection alone does not verify their behavior. See the Service map guide.
3. Create a core-flow test
- Choose Verification → Test list → Generate with AI.
- Select the environment and a use case such as login or basic smoke.
- Review the data and cost notice and check model authentication. A ready local runtime does not mean the model is authenticated. Use the in-app AI connection settings to connect your model. API keys are stored in the OS secure credential store.
- Supply the account conditions and expected behavior, then generate a draft.
- Review its steps, assertions, and execution evidence before applying it. Do not weaken expected outcomes just to obtain a passing result.
Use Verification → Test list for existing tests or imports. You can also write tests yourself; AI is optional. A generated draft and a reviewed, applied test are different stages. Labels above describe the Korean-first desktop UI.
4. Run and review
Open Verification → Test list, select a test and choose Run, and confirm the environment, persona, and browser. In Verification → Run history, inspect failed and skipped items as well as passes. Use failure records, screenshots, and traces to assess the cause. An aggregate pass rate is not release approval.
5. Repeat on the next release
Choose Run this test again (이 테스트 다시 실행) in report details to open run configuration with that test selected. Review the current test, environment, and account before running. This does not restore the code or settings from the historical report.
Run the same tests under the same environment and account conditions, then compare results. In test management, filter by tag and save a view to revisit a set of flows. Check that the saved filter still includes the intended flows before running.
Check that your first run actually tested the expected outcome. On the next release, repeat the test and compare preparation, review, investigation time, and AI costs.
AI data and costs
Execution records are stored on your device, but code, page content, and failure evidence needed for AI tasks may be sent to your configured external provider. Model authentication and provider charges are separate from installation. Conditional fix evaluation may require separate authentication and incur additional charges.
The project session budget is not an overall AI billing cap. Monetary tracking depends on the provider, and failure triage does not use that budget. Avoid AI features for information that must not leave the device. Read more in the FAQ.
Track improvements and changes
Register an improvement from a page or test, recording observed and expected behavior. After a fix, link a new run as evidence and decide its resolution explicitly. Changes compares two runs of structure, screens, or verification results.
After your first check
- Managing UX checks organizes coverage by flow, feature, and persona.
- Reports and analytics explains evidence and recurring failures.
- Scheduled checks repeats prepared tests while the local engine runs.
- MCP collaboration gives external AI access to results and permitted actions.