Write a test with the assistant
Describe the test in plain English; the assistant drafts the steps and checks, you review and confirm, and the tester follows them as written.
Describe it#
In the app, press New test and choose the project (a project is usually one site or app). Choose Web, Android or iOS, then write what the tester should do and what must be true to pass. Name the page it starts at, or several pages: the addresses in your text are found as you type. The first is marked Starts here, the others Also visits, and each shows whether it answers (a quick request from our cloud; the test browser may still get through when it does not).

Below the box are ideas based on your site, such as "Search for a product and check the first result opens a product page". Sign in as… inserts a stored sign-in account by name. Attach a file adds a test file.
Press Draft the plan (or Ctrl/⌘+Enter).
Review the plan#
The plan has:
- Questions, when only you can answer something (for example which account to use). Answer them and press Apply answers, or Skip a question to run with the stated assumption. The test cannot run with a question open.
- Title, Steps (what the tester does, in order) and Checks (what must be true to pass, specific and visible on the screen). You can edit, reorder, add and remove them. A check that is too short or vague is flagged: say the text, value or state that shows it passed.
- Sign-in accounts the test uses, Values to record (read and reported even when no check depends on them) and Assumptions.
- The options line: screen, time limit, video, trackers, mailbox, phone number, endpoints, files. Change opens them: see Test options.
- Show the text the tester receives: the plan as the tester reads it.

Ask for changes in the chat on the left ("also check the price on the cart page") and press Revise. Each draft or revision uses one of your workspace's assistant drafts for the day (40 on the free trial, 300 on paid plans).
The planning review#
Before a plan is saved, a second pass checks that every step and check is something the tester can do and judge on screen. You see "Reviewing the checklist…", then "Reviewed: the checklist covers every step and check". If it finds a problem, it says what; fix the plan, or Save as text anyway and the tester plans from the text at run time.
Save#
Save keeps the test. Save and run keeps it and starts a run straight away. The confirmed plan becomes the test's checklist ("Confirmed by you"): the tester follows it as written.
From an AI agent or the API#
The same assistant is available as plan_test to AI agents and as POST /api/v1/plan in the API: draft, revise with a change, answer questions, then create the test with the plan.