v1.149.0

Compare Source

Features

• Add asyncQueryReasoningEngine to aiplatform v1 API (6332d33)
• Add asyncQueryReasoningEngine to aiplatform v1beta1 API (6332d33)
• Add delete_evaluation_metric to GenAI SDK (6b63480)
• Add evaluation configuration and dataset runs to TuningJob (6332d33)
• Add evaluation metrics and autorater configuration to the AI Platform v1 API (6332d33)
• Add filter and order_by support to list_evaluation_metrics (a7acc97)
• Add multimodal input support and custom output formats to evaluation metrics (6332d33)
• Add OnlineEvaluator API and update Evaluation API (6332d33)
• Add origin to AttributeContext.Request (6332d33)
• GenAI Client(evals): Add allow_cross_region_model to generate_conversation_scenarios to enable cross region support for gemini 3 based model (dc3b639)
• GenAI SDK client(multimodal) - Allow passing dataset ID in addition to full resource name in dataset methods. (2fd8ca0)
• Migrate batch_predictions to agentplatform (53e16ba)
• Model Registry CopyModel BYOSA (6332d33)
• New field CopyModelRequest.custome_service_account for Model Registry CopyModel BYOSA (6332d33)
• Support VeoLoraTuningSpec in the tuning jobs (6332d33)
• Synchronize google/api/expr/v1alpha1 (6332d33)

Bug Fixes

• Add serializer to async_create_session to address 500 error in Agent Engine (AgentServerMode.EXPERIMENTAL). (ffe7fc7)
• deps: Bump litellm cap to >=1.83.7 for additional CVE remediation (3bd0b25)
• GenAI SDK client(multimodal) - Fix Pydantic validation errors when using create_* in some cases (ca6b45e)
• PrivateEndpoint.raw_predict parameter name (3d380ba)
• Removing Python 3.9 support due to EOL (dbdd7b1)
• Unpin litellm upper bound to allow CVE-2026-35030 remediation (e5e6346)
• Update litellm to >=1.83.0 to resolve security vulnerability (ac5a5e4)

Documentation

• Update comments and formatting (6332d33)
• Update doc string to replace Vertex AI with Gemini Enterprise Agent Platform, update method error message to replace Vertex AI with Gemini Enterprise Agent Platform (previously known as Vertex AI), update converter error message to replace Vertex AI with Gemini Enterprise Agent Platform (18d97b6)
• Update SelectiveGapicGeneration usage doc (6332d33)
• Update various comments (6332d33)

v1.148.1

Compare Source

Bug Fixes

• Disable bound token sharing check for agent_engines (04d5092)
• Fix column name inconsistency (history vs conversation_history) across evaluation flows (37f72e5)
• Fix create_session AttributeError for agents without AdkApp (3e85f64)
• Fix NameError when evaluating ADK agents with AgentTool via GenAI Evals SDK (04c5e02)
• GenAI Client(evals): Lazy-load ADK imports in _evals_common.py to avoid top-level ImportError (c378551)
• GenAI SDK client(multimodal) - Remove UpdateMultimodalDatasetConfig and replace with VertexBaseConfig. (f5dc932)
• Strip None fields from agent_data in GenerateLossClusters to prevent INVALID_ARGUMENT errors (f2d73fd)

v1.148.0

Compare Source

⚠ BREAKING CHANGES

• Methods for multimodal datasets now require the full resource name instead of just the dataset ID.

Features

• Add Agent Engine-level configuration for generation_trigger_config. (e73d4e7)
• Add ingest_events method for Memory Bank. (e73d4e7)
• Add memory_types filter to RetrieveMemories (12d90ad)
• Add RetrieveProfiles. (12d90ad)
• Add structured data and context to MemoryRevision. (12d90ad)
• Add structured data and memory type to Memory. (12d90ad)
• Add subset_topics to Memory Bank GenerateMemories (727b8e0)
• Add support for agent gateway in agent engine (9ed3759)
• Add support for keep alive probe in agent engines (ff859e1)
• Adds cancel_query_job to SDK for agent engine long running async tasks. (04f11d1)
• GenAI Client(evals) - add rich HTML visualization for loss pattern analysis (9e7b2b9)
• GenAI Client(evals) - add user-facing generate_loss_clusters with LRO polling and replay tests (28e5f95)
• GenAI Client(evals) - auto-infer metric/candidate and validate inputs for generate_loss_clusters (7142c62)
• GenAI Client(evals) - support loss analysis in EvaluationRun SDK (eb1066b)
• GenAI SDK client - Update client.prompts.create_version to create a prompt and prompt version for a prompt. Recommend create_version instead of create (1e1faa3)
• GenAI SDK client(multimodal) - Auto-generate display name for multimodal datasets if not provided. (9a9e158)
• GenAI SDK client(multimodal) - Make multimodal_dataset optional in create_from_pandas and create_from_bigframes. (dc89de2)

Bug Fixes

• Add bucket ownership verification to prevent bucket squatting in Model.upload() (9feda02)
• GenAI SDK client(multimodal) - Change "name" parameters to refer to resource name instead of dataset ID. (7c04870)
• GenAI SDK client(multimodal) - Fix "type mismatch" warnings in datasets module. (505aa8c)
• Refine Pyink blank line insertion logic and fix range-based formatting regressions. (e382e45)

Miscellaneous Chores

• Release 1.148.0 (1aaeba7)

v1.147.0

Compare Source

⚠ BREAKING CHANGES

• The labels field has been removed from GeminiExample. The tools and safety_settings fields have been turned into lists.

Features

• Add system_labels parameter to Model Garden deploy methods. (a196cda)
• Add the delete method for the Agent Engine Task Store Service (2f2a211)
• GenAI Client(evals): Route gemini-3 model traffic to global region in local agent scraping (e2e81c9)
• GenAI SDK client(multimodal) - Add single_turn_template helper to GeminiRequestReadConfig. (0e5037d)
• Set agent_framework for A2aAgent class. (a8085e5)

Bug Fixes

• Add SDK workaround for double-wrapped Any response in async_retrieve_contexts. (bd4983b)
• GenAI SDK client(multimodal) - Fix inconsistencies between the GeminiExample class definition and API type. (fad250e)

Miscellaneous Chores

• Release 1.147.0 (4a11370)

v1.146.0

Compare Source

⚠ BREAKING CHANGES

• to_bigframes has been removed from the datasets module and moved into the MultimodalDataset class. Instead of dataframe = client.datasets.to_bigframes(multimodal_dataset=multimodal_dataset), use dataframe = multimodal_dataset.to_bigframes() to create a BigFrame instance from a multimodal dataset.

Features

• Add consolidation customization to Memory Bank (a8948c4)
• Add support for custom result parsing in LLM-based evaluation metrics (3e0ddff)
• GenAI Client(evals) - add core data models and code-gen mapping for auto-loss analysis (09794ba)
• GenAI Client(evals): Apply default user persona 'Evaluator' in multi-turn agent scraping (7002dc5)
• In run_query_job, rename gcs_bucket to gcs_uri and allow the case that user sets the filename for the output. (f302d1f)
• Pass state from first bidi_stream_query request to async_create_session (37b5a0f)

Miscellaneous Chores

• GenAI SDK client(multimodal) - Move to_bigframes method to MultimodalDataset class. (6874b8d)
• Release 1.146.0 (aab457d)

v1.145.0

Compare Source

Features

• Add session_id to Create Session to allow custom session id (bea67c2)
• Eval SDK: Migrate model call method by genai SDK usage (ff5e246)
• Eval SDK: Migrate model call method by genai SDK usage in preview foler (ad36123)
• Limit metric registry support to only custom code execution metric and llm based metric in SDK (c12aedc)

Bug Fixes

• Increase default timeout to 600 seconds for ask_contexts and async_retrieve_contexts in VertexRagServiceClient. (3de2c1e)

v1.144.0

Compare Source

⚠ BREAKING CHANGES

• create_from_bigquery and update_multimodal_dataset no longer automatically prepend a missing bq:// prefix for BigQuery URIs. When using the new function MultimodalDataset.set_bigquery_uri the prefix will still be added if needed.

Features

• Add metadata_filter usage to all retrieval and generator methods in rag_retrieval.py (841c597)
• Add container_spec to Reasoning Engine public protos (9a0eefb)
• Add container_spec to Reasoning Engine public protos (9a0eefb)
• Add support for container_spec in AgentEngines (da663c0)
• GenAI SDK client(multimodal) - Add metadata helpers to MultimodalDataset. (e164b19)
• Publish client batch config schema (9a0eefb)
• Refactor evaluation instance building and update LLM metric handler (7a3b436)
• Update the interface for custom code execution metric while maintaining remote_custom_function support for backward compatibility (f7733ec)

Bug Fixes

• Auto-created GCS staging bucket names are less predictable (1a33ad9)
• GenAI SDK client(multimodal) - Replace blocking calls in create_from_pandas with async versions. (2767273)
• Introduce timeout for outgoing A2A requests from AgentEngine (78525d2)
• Relax the requirement to specify class_methods if image_spec/container_spec is provided (6f7b12c)
• Standardize on the app_name in AdkApp (ee9fbe1)

Documentation

• Update API common type documentation (9a0eefb)

Miscellaneous Chores

• Release 1.144.0 (2ad586d)

v1.143.0

Compare Source

Features

• Add AgentEngine Sessions module (368a8f8)
• Add memory_id to Create Memory (2167f36)
• Add RagMetadata and RagDataSchema management APIs (4f0fdfe)
• Add raw_event to Append Event (2167f36)
• GenAI SDK client(multimodal) - Accept gemini_request_read_config instead of template_config in all functions. (f138162)
• GenAI SDK client(multimodal) - Support creating multimodal dataset from bigframe DataFrame (9b7dc29)

Bug Fixes

• Exclude compromised LiteLLM versions from dependencies pin to 1.82.6 (78966da)
• Fix rag resource parsing (bc61708), closes #​6442

v1.142.0

Compare Source

Features

• Add retrieve_contexts_async and ask_contexts SDK methods in rag_retrieval.py (0e0137e)
• Add VALIDATED Function Calling mode to v1 (981a551)
• Add aiohttp to agent_engines dependencies. (394253a)
• Add custom session id field to create session v1 (981a551)
• Add custom session id field to create session v1beta1 (981a551)
• Add EvaluationMetric Create, Get and List methods to Vertex SDK GenAI evals (f4b4244)
• Add model_status to GenerateContentResponse (Gemini API only) (6a03f78)
• Add new embed_content_config to message EmbedContentRequest (981a551)
• Add new embed_content_config to message EmbedContentRequest (981a551)
• Add part_metadata in Part (Gemini API only) (6a03f78)
• Add raw_event field to event proto v1 (981a551)
• Add support for metric_resource_name within metric in rubric generation (4dbd76c)
• Add support for passing agent data for custom code execution metric (0c70de8)
• Add support for referencing registered metrics by resource name in evaluation run API (76a9558)
• Adds run_query_job and check_query_job to SDK for long running async tasks. (0cff2d8)
• Allow using registered metric resource names in evaluation (72942a4)
• Enable a2a streaming for agents deployed to Agent Engine. (ccfd37f)
• GenAI Client(evals) - BREAKING_CHANGE: The agent engine resource name is now passed as a separate agent parameter to create_evaluation_run methods, rather than being part of the AgentInfo object. This parameter is now required if agent_info is provided (dab185a)
• GenAI Client(evals): Add starting_prompt, conversation_plan, agent_data to evaluation datset (64ef223)
• GenAI Client(evals): Update to generate_conversation_scenarios from generate_user_scenarios (1b4c325)
• GenAI SDK client(multimodal) - Support creating multimodal dataset from pandas DataFrame. (8f4b4fe)
• GenAI SDK client(multimodal) - Support getting a bigframe DataFrame from a multimodal dataset (50d804f)
• RAG - Remove VectorDb defaulting in SDK (39b7fe5)
• RagMetadata and RagDataSchema concepts and Batch API definitions added for Preview (981a551)
• Refactor RegisteredMetricHandler and implement llm_based_metric_spec support (24f626c)
• Support rendered_parts in GroundingSupport (6a03f78)
• Update the schema proto for the RagChunk with file_id and chunk_id to align with the vertex_rag_data.proto (981a551)
• Upgrade protobuf from v25.7 to v31.0 (981a551)

Bug Fixes

• Add embedding_metadata to MatchNeighbor from_embedding. (ffd2cff)
• GenAI Client(evals) - missing response warning when agent_data is present (0d6c54e)
• GenAI Client(evals): Drop empty columns in evaluation dataset (be869ca)
• GenAI Client(evals): Remove prompt filled by starting prompt (55b12a5)
• GenAI SDK client - Add get session call to create session sdk if an immediate success is returned (933debd)
• GenAI SDK client - Breaking Change: Remove the following types from vertexai.types and depend on these types from the Google GenAI SDK via google.genai.types - ComputationBasedMetricType, FunctionResponseScheduling, MediaResolution, Outcome, PairwiseChoice, RubricContentType, Blob, CodeExecutionResult, ExecutableCode, FileData, Part, PartMediaResolution, VideoMetadata, FunctionCall, FunctionResponse, FunctionResponseBlob, FunctionResponseFileData, FunctionResponsePart, PartialArg, ComputationBasedMetricSpec, LLMBasedMetricSpec, PredefinedMetricSpec, RubricGenerationSpec, BleuMetricValue, CustomOutput, ExactMatchMetricValue, PairwiseMetricResult, PointwiseMetricResult, RawOutput, RougeMetricValue, BigQuerySource, GcsSource, ModelArmorConfig (3cd949d)

Documentation

• Updated comments for rag_file_metadata_config across ImportRagFilesRequest and UploadRagFileRequest (981a551)

v1.141.0

Compare Source

Features

• "global" endpoint supports "grpc" transport (b3bae32)
• Add PromptTemplateData to support context and history columns when creating Evaluation run from dataframe (e887a2e)
• GenAI Client(evals): Add candidate name to local ADK agent scraping (79d8e1c)
• GenAI Client(evals): Add validation for agent_data in EvaluationDataset in create_evaluation_run (2b0a98c)
• GenAI Client(evals): onboarding agent_config (agents) in create_evaluation_run (eacc86c)
• GenAI Client(evals): Patch agent config map to agent data (8ba4707)
• GenAI Client(evals): Update to enable agent run functionality in eval management service (7a59738)
• Update match_service proto for new embedding metadata fields (fba5350)
• Update match_service proto for new embedding metadata fields (fba5350)
• Update match_service proto for new embedding metadata fields (fba5350)

Bug Fixes

• GenAI Client(evals) - Add datetime and bytes serialization to _json_serializer (7410b1d)

v1.140.0

Compare Source

Features

• A new field computation_based_metric_spec is added to message .google.cloud.aiplatform.v1beta1.Metric (e5f71de)
• A new field evaluation_run is added to message .google.cloud.aiplatform.v1beta1.EvaluateDatasetRun (e5f71de)
• A new field inference_generation_config is added to message .google.cloud.aiplatform.v1beta1.EvaluationConfig (e5f71de)
• A new field llm_based_metric_spec is added to message .google.cloud.aiplatform.v1beta1.Metric (e5f71de)
• A new field metric_results is added to message .google.cloud.aiplatform.v1beta1.EvaluateInstancesResponse (e5f71de)
• A new field predefined_metric_spec is added to message .google.cloud.aiplatform.v1beta1.Metric (e5f71de)
• A new message ComputationBasedMetricSpec is added (e5f71de)
• A new message LLMBasedMetricSpec is added (e5f71de)
• A new message MetricResult is added (e5f71de)
• A new message PredefinedMetricSpec is added (e5f71de)
• Add traffic_type, tool_use_prompt_tokens_details to message GenerateContentResponse.UsageMetadata (e5f71de)
• Add AskContexts and AsyncRetrieveContexts APIs to VertexRagService (fa610af)
• Add AskContexts and AsyncRetrieveContexts APIs to VertexRagService (fa610af)
• Add async wrappers in the GenAI client for the Agent Engine Task Store Service (199a406)
• Add image_spec to Reasoning Engine public protos (e5f71de)
• Add image_spec to Reasoning Engine public protos (e5f71de)
• Add max_concurrent_active_run_count field to Schedule message (e5f71de)
• Add max_concurrent_active_run_count field to Schedule message (e5f71de)
• Add model_armor_config to GeminiExample (e5f71de)
• Add more fields to the GenAI client for the Agent Engine Task Store Service (7d58adb)
• Add resource_reference to the field name in message .google.cloud.aiplatform.v1beta1.Dataset (e5f71de)
• Add resource_reference to the field name in message .google.cloud.aiplatform.v1beta1.DatasetVersion (e5f71de)
• Add support for agent_config_source (ADK) in AgentEngines (204e5b4)
• Add the GenAI client for the Agent Engine Task Store Service (b54ce6d)
• Add transcription fields to Session Service (e5f71de)
• Add v1 API version for Session Service (e5f71de)
• Add Vertex Multimodal Datasets as input and output types in google.cloud.aiplatform.v1beta1.BatchPredictionJob (e5f71de)
• Add Vertex Multimodal Datasets as input and output types in google.cloud.aiplatform.v1beta1.BatchPredictionJob (e5f71de)
• GenAI Client(evals) - Add 3 new multi-turn predefined metrics for agent evaluation (MULTI_TURN_TOOL_USE_QUALITY, MULTI_TURN_TRAJECTORY_QUALITY, MULTI_TURN_TASK_SUCCESS). (513d4d6)
• GenAI Client(evals) - Refactor evaluation visualization to support multi-agent traces and system topology. (84c5602)
• GenAI Client(evals): Add local ADK agent multi-turn agent scraping via ADK user simulation library (429a182)
• GenAI Client(evals): Add user scenario generation function to eval components (87a5125)
• Update data types from discovery doc. (6b5cc8f)

Bug Fixes

• Add safeguards and warnings for remote code execution during pickle-based model deserialization (aa22f54)
• GenAI Client(evals): Avoid required session_inputs column to eval dataset (134b989)
• Only include CandidateResponse if a response is present (10efc13)
• Save artifact in streaming agent run with events when multiturn (e9f1c88)

Documentation

• A comment for enum AggregationMetric is changed (e5f71de)
• A comment for field aggregation_output in message .google.cloud.aiplatform.v1beta1.EvaluateDatasetResponse is changed (e5f71de)
• A comment for field expire_time in message .google.cloud.aiplatform.v1beta1.Session is changed (e5f71de)
• A comment for field filter in message .google.cloud.aiplatform.v1beta1.ListSessionsRequest is changed (e5f71de)
• A comment for field output_info in message .google.cloud.aiplatform.v1beta1.EvaluateDatasetResponse is changed (e5f71de)
• A comment for field ttl in message .google.cloud.aiplatform.v1beta1.Session is changed (e5f71de)
• A comment for message EvaluateDatasetResponse is changed (e5f71de)
• A comment for message Metric is changed (e5f71de)
• A comment for message OutputInfo is changed (e5f71de)
• Comment on evaluate_dataset_response clarified (e5f71de)
• Update documentation of .google.cloud.aiplatform.v1.DatasetService, .google.cloud.aiplatform.v1.Dataset, and .google.cloud.aiplatform.v1.DatasetVersion (e5f71de)
• Update documentation of .google.cloud.aiplatform.v1beta1.Dataset and .google.cloud.aiplatform.v1beta1.DatasetVersion (e5f71de)

v1.139.0

Compare Source

Features

• Add max_concurrent_active_run_count support to Scheduled Pipelines client GA. (a204e74)
• GenAI Client(evals) - update SDK type definitions for Agent Data (6ac28a5)

v1.138.0

Compare Source

Features

• Add support for BYO-dockerfile in AE deployment (7572601)
• GenAI SDK client - Make operation polling interval configurable when creating agent engine sandbox (bf9e0ff)
• GenAI SDK client(multimodal) - Support Assess Batch Prediction Resources for the multimodal datasets. (0fe5314)
• GenAI SDK client(multimodal) - Support Assess Batch Prediction Validity for the multimodal datasets. (a63e8d5)
• GenAI SDK client(multimodal) - Support Assess Tuning Validity for multimodal dataset. (12f5aa5)
• Update the ADK template to export logs directly to Cloud Logging when OTEL_SEMCONV_STABILITY_OPT_IN is set to "gen_ai_latest_experimental". (82db4ad)

Bug Fixes

• Refactor session retrieval fallback in _streaming_agent_run_with_events. (8aec754)

v1.137.0

Compare Source

Features

• Add filter_groups to PurgeMemories for metadata-based filtering (6907f89)
• Deprecate prompt_optimizer.optimize and prompt_optimizer.optimize_prompt in favor of prompts.launch_optimization_job and prompts.optimize (ff811f5)

Bug Fixes

• Use GOOGLE_CLOUD_AGENT_ENGINE_LOCATION env var for service locations. (04aacbb)

v1.136.0

Compare Source

Features

• Add fps to message VideoMetadata (157381a)
• Add agent_card to agent engine spec (d685d81)
• Force-delete for RagCorpora, ignoring any external errors and ensuring deletion of the RagCorpus (157381a)
• Update data types from discovery doc. (a5748fd)

Bug Fixes

• Support custom credentials in RAG.upload_file (66c4d85), closes #​4986
• Test fix internal (1bbf7bb)

v1.135.0

Compare Source

Features

• Add ComputationBasedMetricSpec to support metrics like EXACT_MATCH, BLEU, and ROUGE in EvaluationRun configurations. (9d32dd5)
• Add autoscaling_target_dcgm_fi_dev_gpu_util, autoscaling_target_vllm_gpu_cache_usage_perc, autoscaling_target_vllm_num_requests_waiting options in model deployment on Endpoint & Model classes. (0179aa5)
• Add optimize and launch_optimization_job methods to prompts module (044c3fa)
• List all Model Garden models (54260fd)
• RAG - Add Serverless and Spanner modes in preview. (79da831)
• Remove experimental warning from sandboxes and add specific warning for non-code execution sandboxes. (7d31d13)
• Update the v1 service definition to add embedding_metadata. (8f5bfc5)

Documentation

• Update prompt optimizer code in readme to use methods from prompts module (4636507)

v1.17.0: google-cloud-trace 1.17.0

Compare Source

1.17.0 (2025-10-20)