Gemini Enterprise Agent Platform: Updates and Deprecations

The Grok 4.1 model family, including xai/grok-4.1-fast-reasoning and xai/grok-4.1-fast-non-reasoning, is deprecated on the Gemini Enterprise Agent Platform. These models will be shut down on August 20, 2026, after which API requests to the chat completions endpoint using these model IDs will fail. Users are advised to migrate to newer xAI models like Grok 4.2 or 4.3, or alternative models from the Google Cloud Model Garden.

The Memory Bank IngestEvents API is now generally available, decoupling event ingestion from memory generation. This allows continuous streaming of content and configurable triggering of memory generation. Key features include carrying over context between generation windows using overlap_event_count, configuring memory revisions with revision_labels, revision_ttl, or disable_memory_revisions, and attaching metadata using metadata and metadata_merge_strategy parameters. Details are available in Ingest events.

Memory Bank now supports the gemini-embedding-2 model for similarity search configurations. When using this model, only global, us, or eu endpoints are supported; regional locations are not. Refer to Similarity search configuration for more details.