With the proliferation of networked sensors and artificial intelligence, there is an increasing need for edge computing where data is processed at the sensor level to reduce bandwidth and latency while still preserving energy efficiency. In this talk, I will discuss how meta-optics can be used to implement computation for optical edge sensors, serving to off-load computationally expensive convolutional operations from the digital platform, reducing both latency and power consumption. Meta-optics can also take advantage of additional information channels, such as polarization, spectral composition, and angle of incidence, to process information not recorded on conventional cameras. I will discuss how this increased freedom in design allows meta-optics to augment, or replace, conventional imaging optics in achieving parallel optical processing across multiple independent channels. The meta-optic frontend is demonstrated to enable segmenting and classifying objects with minimal computational resources as well as allow for spectral discrimination in classifying objects based on their emission characteristics.