Large language models are stateless by design, which makes chatbots feel forgetful without proper context management. Spring AI's chat memory support solves this by storing conversation messages and injecting them into each new prompt. The post covers the ChatMemory abstraction, MessageWindowChatMemory for sliding-window context, and how MessageChatMemoryAdvisor connects ChatClient to memory. It walks through setup with in-memory storage for development, then migrating to PostgreSQL via JDBC for persistence across restarts and multiple instances. Practical patterns include using UUID cookies to isolate per-browser conversations, schema initialization options, and production recommendations like Flyway migrations and separating chat memory from full chat history.
