There are several ways to encode λ-terms in interaction nets; for instance, using the original optimal algorithm by Lamping, or compiling λ-calculus into interaction combinators. However, all the encodings we know implement only β-reduction, and not extensional λ-calculus (with η-reduction as well).
Is it possible to encode λ-terms into interaction nets implementing βη-reduction?
(The normal form of a net for a term should correspond to the βη-normal form of the term if any.)