This Vertex AI v1beta1 update introduces new breaking changes by making `task` and `parent` parameters strictly required for specific agent task methods. However, it also rolls back previous breaking changes by making `Agent.status` and `AgentStatus.state` writable again. Key new features include Omni LoRA tuning for advanced model customization, expanded OAuth scopes for Discovery Engine integration, and new agent task management capabilities.
Vertex AI has reversed a recent update, making `Agent.status` and `AgentStatus.state` properties writable once more, restoring developer control. Additionally, many `operations.get` methods now support new Discovery Engine-related OAuth scopes, indicating deeper integration with Agentspace, Vertex AI Search, and NotebookLM Enterprise.
This Vertex AI v1beta1 update introduces significant breaking changes, making several parameters required and many properties read-only. Developers must adapt to the removal of the `media.upload` method and the retraction of recently added `idpNonce` and `ids` parameters. On the feature side, the API sees a massive expansion in GenAI capabilities, including detailed interaction tracking, new content types, and comprehensive agent task management, alongside the re-introduction of `continuationToken` for long-running GenAI operations.
This update introduces breaking changes by making several agent-related schema properties read-only and removing recently added parameters. On the feature side, Vertex AI Agents gain significant new configuration options for identity, memory, encryption, and persona. A new method allows setting default Notebook Runtime Templates, and GenAI models get new capabilities like Bash code execution and video upscaling.
This update brings significant breaking changes, including the removal of `evaluationTasks.operations` and a rollback of `continuationToken` in `GenaiVertexV1beta1Interaction`. Developers must update their code. On the feature side, new file management and batch embedding capabilities are now available, alongside enhancements to video generation.
This update introduces several breaking changes by making key parameters strictly required and certain schema properties read-only. Developers also gain significant new file management capabilities for publisher models and a new batch embedding method.
This Vertex AI v1beta1 update introduces significant breaking changes, making several properties read-only and altering a schema type. On the feature front, you can now resume long-running content generation with new continuation tokens and access detailed content metadata, including citations and grounding sources.
This update introduces significant breaking changes by making several core parameters strictly required and numerous `GenaiVertexV1Interaction` properties read-only. On the feature front, a massive expansion of the `interactions` API enables advanced streaming, tool use (like code execution, file search, Google Maps/Search), and rich content handling for generative AI applications.
Vertex AI's v1beta1 API now offers powerful new capabilities for video generation, including LoRA for stylization and Pro Super Res upsampling. Agents gain more control with default model configurations, and evaluation runs can now utilize cross-region models.
Vertex AI introduces LoRA and Super Resolution configurations for large vision model video generation experiments. Agents can now specify a default model and temperature, and evaluation runs gain cross-region model flexibility.
Vertex AI v1beta1 introduces custom voice management, allowing creation and listing of personalized voices. It also adds granular cache configuration updates. Be aware of several breaking changes, including new read-only fields and a type change for `structuredPrompt` in video generation. A previous breaking change regarding `A2ATaskRecord` read-only fields has been rolled back.
Vertex AI introduces a new `ServingProfile` resource for managing model serving configurations. This update also includes several breaking changes, notably making `parent` and `name` parameters strictly required, and changing the type of `structuredPrompt` in vision models. OAuth scopes for `agents` methods have also been adjusted.
Vertex AI v1beta1 shifts Agent updates to Long Running Operations and introduces experimental 'Pro Edit' video capabilities. New observability features and Reasoning Engine garbage collection also debut.
Vertex AI introduces experimental video editing capabilities and OpenTelemetry-based observability for Agents. Notably, updating an Agent via patch is now a long-running operation.
Vertex AI v1beta1 enforces mandatory name and parent parameters and transitions many resource fields to read-only status. Additionally, the API is shifting its primary OAuth scope to the service-specific aiplatform scope.
Vertex AI is promoting the specific `aiplatform` OAuth scope as the primary authorization method across most resources, while finalizing the enforcement of the `name` parameter as strictly required.
Vertex AI v1beta1 introduces several breaking changes, including mandatory resource parameters and the removal of the recently added transcription 'mode'. It also adds standard IAM policy management for Agents and new sandbox authorization methods.
Vertex AI introduces IAM support and enhanced deletion controls for Agents, alongside significant breaking changes to Sandbox networking and Audio Transcription configurations.
Vertex AI adds a 'Smart' transcription mode for automated cleanup of audio transcripts and introduces multi-speaker configurations. Video generation is enhanced with a new 'Extend' task and agentic processing modes for dynamic frame navigation.
Vertex AI introduces lifecycle management for endpoint responses and a new "Smart" mode for audio transcription that automatically cleans up disfluencies.
Vertex AI introduces advanced video generation capabilities including outpainting and audio control, while refining agent monitoring filters and deprecating legacy ASR fields.
Vertex AI expands its video generation experimental suite with new capabilities for outpainting, video-to-video transformations, and granular audio track management.
Vertex AI introduces session compaction for long-running agent histories and standardizes speech transcription configurations, alongside a breaking change to traffic type enums.
Reasoning Engines now support session history compaction to manage context windows via summarization. Audio transcription APIs are being streamlined with new language code fields, and evaluation runs gain CMEK support.
Vertex AI expands agentic capabilities with Task Stores for decoupled task management and integrated speech-to-text transcription for generative models.
Vertex AI introduces a 'compact' HTTP forwarding method for publisher endpoints and adds standard long-running operation management for Serving Profiles.
New event ingestion and profile retrieval methods for Memory Banks expand agentic capabilities, while breaking changes remove several fields from Agent and Tool result schemas.
Vertex AI introduces native audio transcription schemas and expands agent memory capabilities with event ingestion. New controls for video generation rewriters and granular index endpoint updates are also now available.
This update introduces significant breaking changes by removing generation penalties and caching fields, while adding new proxy methods for model responses and expanding retrieval tool support for agents.
Vertex AI introduces a flexible HTTP forwarding method for publisher endpoints and adds experimental alignment controls for video generation and grounding observability.
Vertex AI expands its Reasoning Engine capabilities with full CRUD support for Memory Banks and introduces a distinction between legacy and step-based streaming events.
Vertex AI expands its memory management capabilities, allowing developers to create, list, and automatically generate persistent memories for AI agents within the Reasoning Engine framework.
Vertex AI introduces MemoryBanks and Memories for persistent agent state management, alongside schema refinements for Monitored Agents and user scenario generation.
Vertex AI introduces long-running operation management for Memory resources and updates Agent scenario generation to support Gemini-based configurations.
Vertex AI adds prompt injection detection for computer-use tasks and Zero Data Retention for Parallel AI Search, alongside significant removals of Agent-related methods.
Vertex AI has significantly expanded its v1 API surface, introducing dedicated resources for managing AI Agents and a new Semantic Governance Policy Engine.
This update introduces Semantic Governance Policy Engines and Agent Anomaly Detection while introducing several breaking changes to the Agents and Interactions APIs.
Vertex AI adds full operation management for Online Evaluators and marks several core GenerationConfig fields as deprecated in favor of a unified response format.
Email Alert Preferences
Stay notified of breaking changes and schema evolutions
Instant Breaking Alerts
Email notification within 1 hour of any removed method or breaking schema evolution.
Weekly Monday Digest
Curated summary of new APIs, beta promotions, and high-impact updates.
Zero Spam Guarantee
1-click unsubscribe anytime. Only verified Google accounts.
