Skip to main content
Momentic uses several specialized AI agents. Each agent is independently versioned so you can upgrade at your own pace and test changes before rolling them out.

​
Configure

Set the version of each agent under ai.agentConfig in momentic.config.yaml:
momentic.config.yaml
ai:
  agentConfig:
    locator: v3
    assertion: v3
    visual-assertion: v3
    text-extraction: v3
v3 is recommended for new projects. Omit an agent to use the organization default.

​
Agents

AgentPowers
locatorClick, Type, and Element check steps
assertionAI check steps
visual-assertionVisual assertions against a screenshot
text-extractionAI extract steps

​
Locator

Locates elements on the page from a natural language description.
  • v1: 2024 models. Lenient matching, picks candidates that fulfill the same function as the description.
  • v2: early-2025 models. Strict about relative positioning (“the link in the header”) and single-quoted text (“the ‘Submit’ button”).
  • v3 (recommended): latest 2025 models. Smarter caches that store referenced attributes and relative elements, so meaningful changes invalidate the cache. Writes clearer reasoning for chosen elements.

​
Assertion

Evaluates natural language statements against a snapshot of the page.
  • v1: 2024 models. Lenient, evaluates statements as “overall true/false”.
  • v2: early-2025 models. Stronger at sorted order, visual positioning, and relative criteria. Strict about single-quoted text.
  • v3 (recommended): latest 2025 models.

​
Visual assertion

Evaluates natural language statements purely from a viewport screenshot.
  • v1: 2024 models. Faster (~3s avg). Lenient on color assertions.
  • v2: early-2025 models. Slower but sharper on small visual details, logical reasoning, negations, and container-scoped assertions. Strict about “exactly” and single-quoted text.
  • v3 (recommended): latest 2025 models.

​
Text extraction

Extracts structured data from the page given a JSON schema.
  • v1: 2024 models.
  • v2: early-2025 models. Tighter schema adherence; throws if extracted data doesn’t conform. Understands the pattern JSON schema option.
  • v3 (recommended): latest 2025 models. Handles nested objects and arrays. Better at following transformation instructions before returning data.