We explore the circumgalactic medium (CGM) of two simulated star-forming galaxies with luminosities L ≈ 0.1 and 1 L⋆ generated using the smooth particle hydrodynamic code gasoline. These simulations are part of the Making Galaxies In a Cosmological Context (magicc) program in which the stellar feedback is tuned to match the stellar mass–halo mass relationship. For comparison, each galaxy was also simulated using a ‘lower feedback’ (LF) model which has strength comparable to other implementations in the literature. The ‘magicc feedback’ (MF) model has a higher incidence of massive stars and an approximately two times higher energy input per supernova. Apart from the low-mass halo using LF, each galaxy exhibits a metal-enriched CGM that extends to approximately the virial radius. A significant fraction of this gas has been heated in supernova explosions in the disc and subsequently ejected into the CGM where it is predicted to give rise to substantial O vi absorption. The simulations do not yet address the question of what happens to the O vi when the galaxies stop forming stars. Our models also predict a reservoir of cool H i clouds that show strong Lyα absorption to several hundred kpc. Comparing these models to recent surveys with the Hubble Space Telescope, we find that only the MF models have sufficient O vi and H i gas in the CGM to reproduce the observed distributions. In separate analyses, these same MF models also show better agreement with other galaxy observables (e.g. rotation curves, surface brightness profiles and H i gas distribution). We infer that the CGM is the dominant reservoir of baryons for galaxy haloes.