Developed and delivered a feature for the langchain4j/langchain4j repository to enhance observability of cache token usage in chat responses. This work involved extending the existing TokenUsage class with a new BedrockTokenUsage implementation, enabling precise tracking of cached token metrics within ChatResponse objects. The approach included designing and integrating new Java classes, writing comprehensive unit and integration tests to ensure accuracy, and preparing the foundation for future documentation. By focusing on API and backend development in Java, the contribution improved cost awareness and data-driven optimization for token usage, supporting more transparent and efficient chat model operations without addressing bug fixes.