Show HN: Jurl – curl that reads the page for you

jurl is a command-line tool like curl that "reads" web pages and extracts exactly the parts you ask for: the gist, a single answer, code blocks, links, images, or photos matching a description. Commands include simple forms (jurl url, jurl -q "question", jurl code, jurl links, jurl image, jurl find "…"), and output is plain text or plain URLs so results pipe cleanly into other tools. It handles JavaScript-driven sites by rendering them with a bundled headless browser (Lightpanda). Typical runs are sub-second and extremely cheap (about $0.0004-$0.0013 per page for text; a few cents for image searches), and examples show it returning install commands, tutorial paragraphs, lists of links, or the best-matching photos from pages with many images.

Under the hood it doesn’t generate text: every candidate paragraph, link or image is scored by decision models - Jev reads text, Clef inspects pixels - and jurl prints winners verbatim when they clear a confidence threshold. Requests go directly to TypeSafe (text) and Cloudflare (vision) using the user’s API keys; there’s no central jurl server or account, and no cookies are sent so logged-in pages are inaccessible. Installation is a two-line script, Homebrew, cargo or PowerShell; source and most components are MIT/Apache-2.0 (Lightpanda is AGPL-3.0).