Let your AI assistant test on real devices

Yes. Paste a requirement, or point at a linked Jira, GitHub or GitLab issue, and generation returns up to ten structured cases with steps, priority and expected results. They arrive as an editable draft you review and keep, not a finished suite that appears on its own.

No. The 11 Insight reports — flaky tests, risk heatmaps, release readiness and the rest — are calculated from your own execution history with queries and heuristics. No model is involved and they spend no AI credits. They are part of 37 reports included on every plan. We used to call them AI-powered; they are not, so we stopped.

Generation spends AI credits, because a model call has a real cost. Paid plans include 100 credits per user per month, reset at the start of the month and not rolled over. A free account gets 100 once, for the life of the account, so you can decide whether the output is worth it. Reporting spends no credits at all.

No. The agent reads and changes test cases, cycles, plans and executions, and records results — the bulk edits that are tedious by hand. Running things on real devices is what Live and Automate do. The value of having both in one product is that the cases the agent writes and the results it reads come from real devices in the same workspace.

That depends on what you hand it. Given a feature name, a model produces cases anyone could have guessed; given real acceptance criteria — token expiry, rate limits, what happens to an existing session — it produces the ones people forget to write. Because everything generated is an editable draft, you keep the cases worth running and drop the rest.

No. Generation runs only when you invoke it, on the specific requirement you choose. Nothing is generated in the background, and the reporting you would use daily involves no model at all.