Resources
AI configuration
Customize AI behavior in your tests
Momentic uses a variety of agents, each specialized for a specific task within end-to-end testing. Currently, the following agents can be configured by users:
Momentic exposes the ability to choose the versions of these agents so that users can upgrade at their own cadence and test upgrades before applying them.
| Name | Description |
|---|---|
| Locator | Locates elements on the page based on a natural language description. This agent is used for all interactive commands that target elements, such as Click and Type steps. This agent is also used to resolve elements in Element check steps. |
| Assertion | Evaluates the truthiness of natural language statements based on an instantaneous snapshot of the page. This agent is used to power AI check steps. |
Locator
v1
- Uses older 2024 models.
- Interprets the user’s criteria more leniently and will attempt to find a candidate element that fulfills the same function as what the user describes.
v2
- Recommended configuration for new organizations.
- Uses the latest 2025 models.
- Places a strong emphasis on respecting relative positioning criteria (e.g. “the link in the header”, “the X icon in the modal”), and will fail if it cannot find the reference element.
- Strongly respects single-quoted text (e.g. “the ‘Submit’ button”) and will reject candidate elements that do not contain that exact text.
Assertion
v1
- Uses older 2024 models.
- Interprets assertion criteria more leniently and will attempt to evaluate whether or not an assertion is “overall true” or “overall false”.
v2
- Recommended configuration for new organizations.
- Uses the latest 2025 models.
- Displays superior understanding of sorted order, visual positioning, and relative positioning criteria (e.g. “The button in the modal is enabled”).
- Strongly respects single-quoted text (e.g. “element with title=‘hello world’”) and will fail the assertion if there is no element that contains that provided text.
Visual Assertion
v1
- Uses older 2024 models.
- Faster, returning results in an average of 3 seconds.
- More lenient when evaluating user criteria against the page state, specifically for assertions about color.
v2
- Recommended configuration for new organizations.
- Uses the latest 2025 models.
- Slower than V1 but perceives small visual details more accurately.
- Far stronger on logical reasoning, negated assertions, and assertions about elements in containers.
- Strongly respects the term “exactly” and single quotes around text terms. Will fail the assertion if it not unambiguously true when these modifiers are provided.
Text Extraction
v1
- Uses older 2024 models.
v2
- Recommended configuration for new organizations.
- Uses the latest 2025 models.
- Adheres to the provided JSON schema more consistently and will throw an error if the extracted data does not conform to the schema.