Machine-learning-enabled atomistic simulations, spectroscopy, electrochemical data and quantum-chemical calculations reveal previously unstudied multi-lithium-ion complexes without bound anions, improving understanding of lithium-ion-battery functionality at the atomic level.