OCI GoldenGate brings embeddings into replication

GoldenGate is evolving from a data-movement layer into AI data infrastructure, eliminating a separate post-replication embedding pipeline. The tradeoff is that teams must carefully manage embedding latency, model consistency, costs, and failure handling inside their replication architecture.

• –Inline embeddings keep RAG and semantic-search indexes synchronized with operational data.
• –AI Model connections support multiple providers, reducing dependence on Oracle’s model stack.
• –The `@AISERVICE` function lets developers add embeddings directly to Replicat mappings.
• –Model version and vector-dimension changes require migration discipline to avoid semantic drift.
• –Embedding generation during CDC may increase replication latency and external API costs.