How dejav works

Describe

Give dejav a description of what your app should do: bullet points, prose, or cells pasted from a spreadsheet. No particular format is required.

If your specs are in a GitHub repository, point the project at their folder. dejav reads the files when a scan starts and deletes its copy when the scan ends.

Generate

A scan reads your description against your site's pages and writes test cases. Each case has a title and a description of what to check and what should happen, and most name the page they start on. The intermediate results are saved with the scan, so you can check how each case was derived.

For each feature, dejav also writes cases you did not list. [Edge] cases cover boundary values, double submits, going back mid-flow, and empty states. [Error] cases cover invalid input, XSS and SQL injection, and access without signing in.

Learn

Choose Learn, and the AI reads each test case and its page and works out concrete steps. A check run plays them in a sandboxed browser session to confirm they pass.

The steps are saved with the test case. Only a test case whose check run passes is marked learned. When the check run does not pass, the learning job shows the reason for that test case.

Replay

Once a test case is learned, runs replay its learned steps without AI and still record step results, screenshots, and a session replay.

A test case found by exploring your site can also run before it is learned. It runs again along the route it was found on, with AI checking the results.

Relearn

When a UI change breaks a learned test case, the run reports it as failed and the test case is marked as needing relearning. Choose Relearn, and dejav learns it again and plays the new steps in a check run. If the check run fails, the test case keeps its earlier steps and stays marked as needing relearning.