The design of new host–guest complexes represents a fundamental challenge in supramolecular chemistry. At the same time, it opens new opportunities in material sciences or biotechnological applications. A computational tool capable of automatically predicting the binding free energy of any host–guest complex would be a great aid in the design of new host systems, or to identify new guest molecules for a given host. We aim to build such a platform and have used the SAMPL7 challenge to test several methods and design a specific computational pipeline. Predictions will be based on machine learning (when previous knowledge is available) or a physics-based method (otherwise). The formerly delivered predictions with an RMSE of 1.67 kcal/mol but will require further work to identify when a specific system is outside of the scope of the model. The latter is combines the semiempirical GFN2B functional, with docking, molecular mechanics, and molecular dynamics. Correct predictions (RMSE of 1.45 kcal/mol) are contingent on the identification of the correct binding mode, which can be very challenging for host–guest systems with a large number of degrees of freedom. Participation in the blind SAMPL7 challenge provided fundamental direction to the project. More advanced versions of the pipeline will be tested against future SAMPL challenges.