Described herein are systems, devices, and methods for content delivery on the Internet. In certain non-limiting embodiments, a caching model is provided that can support caching for indefinite time periods, potentially with infinite or relatively long time-to-live values, yet provide prompt updates when the underlying origin content changes. Origin-generated tokens can drive the process of caching, and can be used as handles for later invalidating origin responses within caching proxy servers delivering the content. Tokens can also be used to control object caching behavior at a server, and in particular to control how an object is indexed in cache and who it may be served to. Tokens may indicate, for example, that responses to certain requested URL paths are public, or may be used to map user-id in a client request to a group for purposes of locating valid cache entries in response to subsequent client requests.