The document presents RL-Cache, a reinforcement learning-based cache admission system for content delivery networks (CDNs) aimed at maximizing cache hit rates in edge servers. It details the challenges of caching, introduces the application of direct policy search for training the model, and illustrates the results from evaluations using real-world CDN data. The findings indicate that RL-Cache can outperform existing caching algorithms and adapt effectively across various traffic classes and cache sizes.