Decisium

From behavioral specification to intelligent execution.

Semantic UI testing with decision models. Decisium runs Gherkin features written in plain English against a real browser. You write When I schedule the campaign, not selectors.

Each Step is executed by Jev, TypeSafe's decision model. Jev does not generate text or code: it picks among the options Decisium enumerates from the page (the elements, the actions, the check kinds) and returns a probability for each. Decisium's code, not the model, decides whether a Step passed. No LLM, no prompts, no hallucinated selectors. See How a Step runs for the questions it is asked and how many requests a Step takes.

@url:https://shop.example.com
Feature: Checkout

  Scenario: Buy a backpack
    When I add the "Backpack" to the cart
    And I check out with email "ada@example.com"
    Then the order status should be "Confirmed"

• No selectors. Steps name what the user wants. Jev picks the element by meaning, among the ones on the page.
• Record and replay. What each Step resolved to is saved in a cassette next to the feature. CI replays it with zero model calls, and a real UI change fails as UI_CHANGED.
• Built on cucumber-js and Playwright. Your cucumber configuration, formatters, profiles and step definitions keep working.
• Agent-ready failures. A failed run writes report.md: the Step, the model's answers, the page text and a rerun command, ready for a coding agent.

Next: Installation, then the Quick start.