momentic
Momentic is an end-to-end testing platform for web, iOS and Android. Its agents turn natural-language instructions into repeatable tests stored as YAML files in a codebase, then run and maintain them. A separate Mo workflow tests a running application and reports reproduced bugs without requiring a test-file setup.
The product is relevant when a team needs to verify user flows after code changes, including code produced by an agent. A generated test suite still needs a defined expected outcome. Passing a test establishes what its assertions checked, rather than proving that the whole application is correct.
Inspect the assertions as well as the actions
Momentic supports natural-language element descriptions and assertions across text, visuals and the DOM. The documentation says cached steps can skip an AI call while their locators still match. Tests remain readable artifacts that a team can version and inspect.
Choose flows whose failure matters to the product, then verify that the assertions cover the resulting state. A checkout test should inspect the total and order result, for example, rather than only confirming that a button was clicked. Use a controlled test account and environment before allowing a test to modify data.
The publisher describes self-healing tests when an intended interface change occurs. Review the resulting diff: a repaired locator should preserve the expected behavior instead of weakening an assertion to make the run green. Session replays and failure traces can help distinguish a product regression from an environment problem.
Make a CI check enforceable
The GitHub Actions guide runs the same repository tests against a pull-request branch. Authentication uses a Momentic API key stored as a repository secret. A failed test returns a nonzero exit code.
The guide distinguishes an advisory job from a required branch-protection check. Adding the check to required status checks is what blocks a merge on failure; a workflow file alone does not enforce that policy. Labeling critical flows can keep the gate focused, while uploaded artifacts preserve evidence for a failure.
Mobile testing also needs the app build associated with the change. The documented upload tags tie an immutable build to the CI run, helping prevent a test from exercising an older application by mistake. Confirm that the test uses the intended platform and build before treating its result as release evidence.
Account for steps and hosted runtime
The pricing page lists two thousand monthly credits on Free and ten thousand on a paid plan at one hundred twenty-five dollars monthly, plus additional usage. It says credits reset monthly without rollover. Estimated test-run counts assume roughly ten steps and are not a fixed run allowance.
Normal steps cost one credit, while AI-generated or failure-recovery steps cost two. Hosted browser and mobile runtime, classification and triage also consume credits. Estimate the complete workflow cost, including retries and investigation, and confirm current rates before expanding a suite.
Start with a small representative set, inspect failures and preserve the artifacts. That produces better evidence than relying on the publisher's aggregate bug counts or selected customer testimonials.