Massive crowd simulation offers valuable insights into human behavior, with applications in urban planning, public safety, computer animation, and beyond. However, traditional methods that treat crowds as individual agents face significant scalability challenges as crowd sizes grow, resulting in performance bottlenecks. While there are advanced animation engines capable of supporting the efficient rendering of large-scale crowds, existing crowd simulation frameworks often face limitations in scalability and flexibility. In response to these challenges, we present TaiCrowd, an open-source crowd simulation framework. TaiCrowd employs a generic parallel approach specifically designed for agent-based methods, filling the current gap for a scalable, efficient, and user-friendly crowd simulation framework. Experiment results show that TaiCrowd achieves a substantial 60-fold improvement for crowd simulations involving hundred thousand individuals compared to other simulation frameworks, outperforming existing frameworks and enabling real-time simulation for large-scale crowd management. TaiCrowd can be accessed from https://github.com/Worter623/TaiCrowd .