Spring AI Session is an event-sourced short-term memory layer designed to replace the older ChatMemory API. It stores conversation history as turn-aware events (user message + all assistant replies and tool calls), ensuring compaction never splits a turn. The tutorial covers setting up the dependency with Spring AI 2.x and Spring Boot 4.x, creating and managing sessions via SessionService, wiring memory transparently into ChatClient using SessionMemoryAdvisor, and choosing among four compaction strategies: SlidingWindowCompactionStrategy, TurnWindowCompactionStrategy, TokenCountCompactionStrategy, and RecursiveSummarizationCompactionStrategy. The first three drop older events cheaply, while recursive summarization uses an LLM call to preserve condensed context. Manual compaction via compact() is also demonstrated. The API is currently incubating in the spring-ai-community project and is slated to become the default replacement for ChatMemory.
