There are 3 use cases: keyed lookups; filtered index reads, and keyed joins.
The keyed lookup will be the equivalent of an index read inside a project transform with a keyed field and optional non-keyed filters. For example, SELF.foo := indx_attr(KEYED(bar=LEFT.my_var) AND extra_field=LEFT.my_extra);
The filtered index read will be the equivalent of a PULL(indx_attr(KEYED(...) AND ...)).
The keyed join will be the equivalent of JOIN(ds, ks, KEYED(...) ..., ...)
The details are covered in the individual JIRA tickets.