AMBER alert systems are inefficient since object searching heavily relies on reports of witnesses, who might miss alerts and cannot search enough areas of city. Using automatic license plate recognition (ALPR) technique, city-wide video surveillance is of great improvement for vehicle searching. However, analyzing huge amount of video data in the cloud leads to vast cost of data transmission and high response latency. Edge computing as an emerging computing paradigm can significantly reduce the cost of data transmission and response latency for latency-sensitive applications due to the data processing at the proximity of data sources. In this poster, we propose an enhanced AMBER alert system using collaborative edges, called AMBER Alert Assistant (A3 in short), which can search the suspect vehicle by analyzing static and mobile cameras' data in real time fashion. We propose location-direction-related diffusion that effectively optimizes the searching area for vehicle searching. The evaluation results show that real-time video analytics can be achieved by collaboratively leveraging multiple edge nodes.