Extended adaptive biasing force (eABF), a collective variable (CV)-based importance-sampling algorithm, has proven to be very robust and efficient compared with the original ABF algorithm. Its implementation in Colvars, a software addition to molecular dynamics (MD) engines, is, however, currently limited to NAMD and LAMMPS. To broaden the scope of eABF and its variants, like its generalized form (egABF), and make them available to other MD engines, e.g., GROMACS, AMBER, CP2K, and openMM, we present a PLUMED-based implementation, called extended-Lagrangian free energy calculation (ELF). This implementation can be used as a stand-alone gradient estimator for other CV-based sampling algorithms, such as temperature-accelerated MD (TAMD) and extended-Lagrangian metadynamics (MtD). ELF provides the end user with a convenient framework to help select the best-suited importance-sampling algorithm for a given application without any commitment to a particular MD engine.