Web tasks often span across multiple webpages and require people to repetitively interact (e.g., click, scroll, copy, type) over them. Because webpages are independently developed and generally lack incentives for interoperability, we envision a client-side remedy. In contrast to end-to-end automation which assumes well-specified information need, we believe that people should have a set of small-step automation primitives to selectively apply in any point of their information work, even with partial information need.

For each webpage, Cling parses its content into a typed, relational database, while maintaining a bidirectional mapping between each database entry and its corresponding DOM element on the page. When people interact with page content, Cling identifies automation opportunities using the type, value, and other metadata of its corresponding database entry, before surfacing and executing them on the original webpage.