Jev in production › Agent tool and action selection
Jev's Choice selects the target page element for plain-English Playwright steps, while a Noul model verifies test claims, cheap enough to run on every pull request.

Write browser tests in plain English, keep the full power of TypeScript, and run the whole suite on every pull request for dollars a month instead of thousands.
Each ai(...) sentence is one step that Sedum resolves on the page, with no selectors. Everything between steps is ordinary code: Playwright's page and expect, test data from any library, API calls to seed or skip setup, and helper functions.
Sedum uses a decision model for bounded tasks such as finding the element a sentence refers to and judging whether a claim holds on the page. TypeSafe Jev is the default; Cloudflare Clef is an opt-in text provider. Both return probabilities, not free text. Everything else is deterministic: clicking, typing, waiting, verdicts, and exit codes run on...
For the project's own README, linking back here: