Runtime Verification (RV) dynamically monitors whether traces—sequences of events like method calls—violate formal specifications. RV is effective at bug finding, but incurs high runtime overheads. Prior work showed that 99.87% of monitors are redundant as they observe the same trace as the other 0.13%. So, we recently proposed a novel approach based on reinforcement learning (RL) to speed up RV by reducing redundant monitoring. We present Valg, a tool that implements that approach for Java. Compared to our prototype, Valg adds (i) per-spec hyperparameters, (ii) RL trajectory saving, (iii) offline hyperparameter tuning, and (iv) performance optimizations. Also, we integrate Valg with the main development branch of JavaMOP and TraceMOP, two state-of-the-art RV tools, and fix a long-standing specification bug. We evaluate Valg on 40 open-source projects, and Valg with per-spec hyperparameters is up to 58.7 percentage points (pp) and 89.7pp faster than our prototype, and Valg checks up to 44.8pp more unique traces. Valg’s offline hyperparameter tuning and our optimizations make JavaMOP, TraceMOP, and Valg faster. Valg is open-sourced at https://github.com/SoftEngResearch/tracemop, and a video demo can be found at: https://youtu.be/_QCyHaa_ICc.