This content introduces ReefWatch, a Coral-powered production triage agent designed to investigate production incidents by querying various systems. It details how the agent discovers connected tools at runtime, correlates evidence across different sources, and provides an inspectable UI for query results.
The article introduces a new 60-line Redis sink for ragvitals, designed to leverage existing Redis instances already present in most production RAG stacks for purposes like prompt or embedding caching. This Redis-based solution offers an efficient alternative to other monitoring sinks for tracking production drift, avoiding separate infrastructure and simplifying data collection.
The article explains that standard CI/CD pipelines miss production breaks in live UCP profiles, such as issues with TLS certificates, CDN propagation, or schema hosts. It advocates for building a UCP watchdog to monitor these dynamic production surfaces and catch problems that don't originate from code commits.