Bioluminescence microscopy offers a uniquely non-invasive window into cellular dynamics, yet its use has traditionally been limited by the intrinsically low brightness of luciferases. The poor photon budget forces long exposures, preventing faithful visualization of rapid physiological processes, especially in three dimensions. To overcome this barrier, we developed a Fourier light field microscope coupled with deep-learning–based reconstruction that achieves sub-second volumetric bioluminescence imaging, enabling direct recovery of 3D volumes from photon-limited measurements while suppressing reconstruction noise. This approach substantially mitigates the speed–resolution trade-off of conventional light field methods and bypasses the need for slow classical deconvolution priors on the depth-dependent point-spread function. We demonstrate its power by performing near real-time 3D calcium imaging in muscles and neurons of freely moving Caenorhabditis elegans , and by quantifying cell dynamics within stem-cell–derived spheroids using fluorescently labeled nuclei. Together, these results establish our framework as a practical tool for dynamic, volumetric studies of living systems.