Although post-combustion reactive absorption from point sources is a leading technology for CO2 capture, currently used solvents are based on first-mile improvements to almost century-old solvent methodology, and to achieve atmospheric CO2 reduction goals, improved solvents are urgently required. The conventional approach for solvent discovery typically entails identifying a prospective solvent by chemical intuition, followed by fitting the parameters of macroscopic thermodynamic models to experimental measurements. We present a novel combined molecular-based and thermodynamic framework augmented by a Machine Learning model that facilitates the large-scale screening of potential solvents at a fraction of the enormous time and cost of the conventional experimental approach. The framework is purely predictive, requiring no experimental data for its implementation, and is based on a typical Henry-based electrolyte solution thermodynamic model for the CO2-loaded solvent in equilibrium with CO2 in its vapour. Its implementation requires values of the equilibrium constants for the solvent-dependent protonation (pKa) and carbamate reversion (pKc) reactions, which are predicted by first-principles molecular-based theory involving quantities that can be calculated using readily available electronic structure and atomistic simulation software. We show their incorporation within open-source Gibbs Energy (GEM) minimization software to calculate the dependence of the CO2-loaded speciation and solubility on the flue-gas partial pressure, P(CO2), solvent composition, temperature and pressure, from which other important properties such as the heat duty and cyclic capacity can be calculated. We also describe the incorporation of Machine Learning models for directly calculating the equilibrium constants from a solvent molecular structure fingerprint to accelerate the molecular-based calculations. We present results validating our methodology, and describe potential extensions for predicting the effects of flue-gas impurities. Although previous researchers have used molecular-based techniques with various levels of accuracy to calculate the equilibrium constants of the deprotonation and carbamate revision reactions involved in the CO2 capture process, our approach is the first to integrate their calculation by means of accurate methodologies with readily available open-source software, yielding a predictive pathway linking a potential solute’s molecular structure to the relationships between its flue-gas CO2 concentration and its solubility and chemical speciation.