Most web scraping fails for an unglamorous reason: it is welded to hand-written CSS selectors, so the moment a site reshuffles its markup the pipeline quietly returns garbage. The wager here is that whoever knows which data matters should not also have to babysit selectors. You either record yourself clicking through a page once, or you describe the fields you want in plain English and let an LLM locate them.
What Sets It Apart
- Two capture modes share one workflow: a recorder that mimics your actions and infers list/table patterns, so a single example row generalizes across a whole paginated list, and an AI mode that takes a natural-language description and skips selectors entirely.
- It is not only extraction. The same tool crawls entire sites, converts pages to clean Markdown/HTML with screenshots, and runs filtered web searches, so "find the pages" and "get the data" live in one place instead of three glued-together libraries.
- Built on Playwright, it copes with JavaScript-heavy pages, logins, and pagination that defeat request-based scrapers, and ships an SDK and CLI so a no-code recording can graduate into a programmatic pipeline without re-platforming.
Who It's For
Great fit if you pull data from dynamic sites and want non-engineers to build and maintain extractors, or you want one self-hosted tool spanning scrape, crawl, search, and AI extraction. Look elsewhere if you need a fully managed cloud service with zero ops, or your targets are simple static pages where a lightweight request-based scraper is cheaper. Note the AGPLv3 license: fine for internal self-hosted use, but it carries copyleft obligations if you fold it into a networked service you distribute.
