Relign is a fully open-source reinforcement learning (RL) library designed to advance AI by realigning base models into reasoning models that can tackle complex problems with human-like, step-by-step logic. Unlike standard models that struggle with reasoning for specific tasks like agent actions, Relign leverages state-of-the-art RL algorithms—such as PPO, with GRPO support on the way—and provides practical abstractions for Chain of Thought (CoT) and Monte Carlo Tree Search (MCTS) inference strategies. This enables developers and researchers to fine-tune models, significantly boosting their reasoning capabilities and, in turn, improving the performance of any system built on top of them. Tailored for creating reasoning engines, Relign supports evaluation on widely recognized reasoning benchmarks, making it a powerful tool for pushing the boundaries of artificial intelligence and its real-world applications.