An Agda implementation of the "maps" representation of lambda terms, as described in:

Viewing λ-terms through maps
Masahiko Sato, Randy Pollack, Helmut Schwichtenberg and Takafumi Sakurai
Indagationes Mathematicae 24 (2013) 1073–1104
http://dx.doi.org/10.1016/j.indag.2013.08.003

This is just the maps representation, but uses Agda's support for induction-induction and recursion-recursion definitions to bypass the symbolic expressions and obtain the representation directly. Also, Agda's support for irrelevance annotations makes the representation canonical: the proofs of the orthogonality of two maps in the representation are marked as irrelevant, which makes proofs of the divisibility relation, and hence the representation of lambda terms, unique.