April 2026 monthly summary for the LangChain4j project focusing on Bedrock prompt caching configurability. Delivered a configurable TTL for Bedrock prompt caching with 5-minute or 1-hour options, while preserving backward compatibility. This enables teams to tailor prompt caching behavior to their API call patterns and cost considerations, improving performance and predictability in response times and metrics collection. Key notes: - Commit 5a0a327c2007af6af691a1d046a99562c3c3dd26 documents the feature addition and rationale. - All cache points (system messages, user messages, tool configurations) now honor a configurable TTL via a dynamic CachePointBlock. - TTL is optional; existing default behavior (5-minute TTL) remains unchanged when TTL is not provided. - TTL is threaded through BedrockChatModel and BedrockStreamingChatModel to extract workflows, preserving metrics and message conversion features. - Backward compatibility was a primary design goal; no changes required for existing user code. Overall impact: - Business value: improves cache control for Bedrock prompts, enabling cost optimization and better caching strategies for diverse workloads. - Technical scope: API surface extended with BedrockChatRequestParameters.cacheTtl and promptCaching(placement, ttl); refactoring across AbstractBedrockChatModel, BedrockChatModel, and BedrockStreamingChatModel to pass TTL through end-to-end. - Technologies/skills demonstrated: Java builder patterns, API design for backward compatibility, integration with AWS Bedrock SDK CacheTTL, cross-module propagation of configuration, and feature documentation.