There are three types of castes in honey bee (Apis mellifera L.) colony: the queen, workers, and drones. Although they are all important to perpetuate the species, drones do not collaborate with tasks in the colony and their counting may overestimate the real number of foraging workers, individuals who really contribute to pollination services. So, monitoring, classifying, and counting the flow and proportion of workers and drones through the beehive entrance provide useful information related to the colony’s well-being. This has become possible thanks to the so-called Precision Beekeeeping, an emerging field of digital agriculture to gather and transfer bee-related data over time. Here, we propose ApisFlow, a real-time object-tracking framework for automatically detecting, tracking, classifying and counting the flow of honey bee castes at the hive entrance. ApisFlow uses computer vision and machine learning methods and algorithms. We strongly believe that ApisFlow allows bee counting, tracking, and classification in a less laborious, safe, fast, and accurate way to help beekeepers in making decisions saving time. Suggesting a high-precision algorithm with a mean error rate below 5%.