Spring AI's Advisors API provides a composable interception layer around ChatClient calls, similar to Servlet filters or AOP advices but scoped to AI request/response cycles. The post walks through building a logging advisor (CallAdvisor) that wraps model calls with before/after logging, and a safety advisor that short-circuits the chain without calling the model when sensitive words are detected. Key concepts covered include advisor ordering (lower getOrder() = higher precedence), the difference between CallAdvisor and StreamAdvisor interfaces, registering advisors as defaults vs. per-request, and a survey of Spring AI 2.0.0 built-in advisors for logging, memory, RAG, safety, tool calling, and output validation. Practical testing and production considerations like redacting sensitive data from logs are also discussed.
