Release archive
3.2.203

RAG-EMBEDDING-LOCATION-SPLIT (AI Runtime)

Release v3.2.203 focuses on RAG-EMBEDDING-LOCATION-SPLIT (AI Runtime).

FixedInternal
RAG-EMBEDDING-LOCATION-SPLIT (AI Runtime)

Release digest

Fixed

  • RAG-EMBEDDING-LOCATION-SPLIT (AI Runtime) — Vertex-backed RAG embeddings now use RAG_EMBEDDING_LOCATION instead of the Gemini model location alias, so Gemini preview rollouts can use GEMINI_LOCATION=global without moving text-embedding-004 lookups away from their regional endpoint.

Internal

  • Added focused regression coverage proving the Vertex embedding service uses the dedicated embedding location even when Gemini model calls are configured for global.
  • Updated the AI runtime architecture notes and deployment environment inventory for the new RAG_EMBEDDING_LOCATION operator contract. Public docs decision: no public documentation change is required because this is an internal runtime/deployment control and does not change user-facing product behavior, legal language, investment guidance, report output, or public API setup.

Technical notes