The Vertex AI RAG Engine has been upgraded in 2026 to support Serverless RAG Mode (public preview) — a fully managed database for RAG that entirely eliminates the need to provision and manage vector databases like Pinecone or Vertex Vector Search manually.
RAG Cross-Corpus Retrieval (public preview): The new AsyncRetrieveContexts API allows a single generative agent to retrieve from multiple corpora simultaneously. For example, an agent can retrieve technical specs from a codebase corpus and pricing data from a PDF corpus in a single operation.
Vector Search 2.0 is now generally available, unifying data and vectors with auto-embeddings. It supports hybrid search combining vector, full-text, and semantic re-ranking in a single query — dramatically simplifying retrieval architectures.
You can now enforce strict schema validations on document metadata, allowing agents to filter vector searches using powerful SQL-like conditions before the semantic search even runs.