Use cases
Six shapes this takes in practice, and the exact call behind each one.
An agent that only fetches HTML is guessing about layout. It can't tell whether the pricing table renders, whether the CTA is above the fold, or whether a consent wall is covering everything. Hand it a viewport capture and it can answer from the picture instead.
{"url":"competitor.com/pricing","mode":"viewport","width":1440}Capture the same routes at the same widths on every deploy and diff the images. No Playwright to maintain, no CI container that needs Chrome, no flake from a runner that ran out of memory.
{"url":"staging.app.com/checkout","mode":"fullpage","width":1440}Prices, badges, hero copy and the position of the signup button change constantly and rarely get announced. A daily full-page capture is a dated record of exactly what a visitor saw.
{"url":"rival.com","mode":"fullpage","waitMs":2000}Terms, disclosures and offers get edited quietly. A PDF is the artifact a reviewer will actually accept: real pages, selectable text, and a fixed date it was taken.
{"url":"vendor.com/terms","mode":"pdf"}Plenty of pages return an empty shell to a plain fetch and fill themselves in client-side. Text mode gives you the rendered DOM as clean text — the version a person would read — ready to embed.
{"url":"docs.somesdk.dev/guide","mode":"text"}Changelog images, launch posts, social cards, docs illustrations. Capture at the exact width you need and skip the round trip through someone's desktop and a cropping tool.
{"url":"app.com/feature","mode":"viewport","deviceScale":2}