To be more specific:
I am trying to estimate a model with two predictors and an interaction feature using an IV. Which would be the most appropriate approach, and how could I implement (the following) model in Grelt?

The model is Y=b0+b1*X1+b2*X2+b3*X1*X2+e
Y is an ordinal six grade Likert variable,
X1 and X2 are binary variables. X1 is endogenous and X2 is exogenous.
Z1 is the Instrumental Variable for X1 and is an ordinal six grade Likert type variable.

If Y is ordinal then a natural additional question would be whether using an ordered probit might be (more) needed.

In that sense I would tend to recommend an explicit two-stage approach, regressing the endogenous RHS terms on the set of valid instruments and continue with the fitted values from there, doing the ordinal probit in the second stage. A standard ordered probit routine is "of course" available in gretl. Just mark the LHS variable as discrete and then gretl will automatically apply the ordered probit estimation.

But I'm no expert for setups such as these, so I certainly also recommend reading up on the textbook econometrics on doing an IV ordered probit estimation. To my knowledge doing ordered probit with IVs is not directly available in gretl, but I may be missing something.

(The HIP addon for gretl does IV-probit, but not IV-ordered-probit, if I understand correctly.)

cheers

sven