Predicting pedestrian trajectories is essential for understanding human behavior and optimizing spatial planning. A key characteristic of pedestrian trajectories is their multimodality, which results from the diverse intentions of individuals. While recent studies have employed various techniques, such as clustering, tree enumeration, and Gaussian mixture models, to address this multimodality, a more natural and efficient approach is to directly model the distribution of trajectories. To address this need, we propose a spatio-temporal aware flow matching framework for pedestrian trajectory prediction. This framework empowers flow matching-based generative models by enabling them to analyze past trajectories of both the subject and their neighbors so as to model the distribution of future trajectories. Benchmarking results demonstrate the superiority of our proposed framework, highlighting its ability to achieve more accurate and efficient trajectory predictions compared to existing methods.