PointCheck is a WCAG 2.1 & 2.2 Level AA accessibility tester built by Brendan Works.

Most automated accessibility tools — Axe, Lighthouse, browser extensions — work by inspecting the DOM. They can tell you if an image is missing alt text or if a button lacks an accessible name. What they can't do is look at the page the way a human eye would. That means they miss failures that only show up visually: a focus ring defined in CSS that gets overridden by a widget, a contrast failure caused by a layered background, or an element only reachable by mouse.

PointCheck uses MolmoWeb-8B — Allen AI's open-source vision-language model trained for web navigation — to seethe browser the way a human would. MolmoWeb takes a screenshot and returns a pixel coordinate, pinpointing exactly where the focused element is on screen. A second model, Molmo-7B-D, then answers a direct question about what's visible in that region: “Is there a visible focus indicator? Describe it.”

This two-model visual pipeline catches a category of failures that DOM-only tools cannot: focus rings that exist in CSS but are visually absent, interactive content that is only reachable by mouse, and contrast failures on elements with composed transparent backgrounds.

The rest of the test suite covers keyboard navigation, 200% zoom reflow, color-blindness simulation, form error handling, and a broad page structure check. After all checks complete, Allen AI's OLMo-3 writes a plain-English executive summary — no accessibility jargon — covering what was found, what failed, and what to fix first.

Paste any public URL, select your tests, and get a detailed accessibility report with visual evidence streamed live.

Read the full technical writeup on Substack →