2.10.0 (2026-11-12)
Features Added
- Promoted Agent Insights monitors, runs, insights, and related models to GA through
AgentInsightMonitorsClientand
AgentInsightMonitorsAsyncClient. Build these clients directly throughAIProjectClientBuilder; the Agent Insights
preview header is no longer required. - Added
ListAgentInsightsOptionsandListAgentInsightRunsOptionsfor filtered paging. Each options object contains
the monitor ID as well as cursors, page size, ordering, and operation-specific filters. - Added monitor priorities through
AgentInsightPrioritiesandAgentInsightPrioritiesDetails, and generated agent
summaries throughAgentInsightSummary. - Added scheduled dataset and trace evaluation rules through
DatasetEvaluationRuleAction,
TraceEvaluationRuleAction,EvaluationRule.schedule, timer-trigger events, and schedule provisioning status/error
reporting. Scheduling trigger and recurrence models are now GA. - Added evaluation run data sources for CSV files, target completions, traces, and user-conversation simulation.
Trace selection supports trace and conversation IDs, agent filters, time ranges, sampling, and ingestion delay.
Added item-generation parameters for response retrieval, synthetic data generation, target completions, and
red teaming with seed prompts and taxonomies. - Added typed
DatasetEvaluationSource,TraceEvaluationSource, andUserConversationSimulationSourceunions with
source-specific factories such asofCsv(...),ofTraces(...), andofFileId(...), respectively, and typed getters.
Standard and preview conversation simulation data sources shareUserConversationSimulationSource. Evaluation
sources reuse OpenAI types directly where applicable; simulation scenarios and evaluation-model settings reuse
Agents types. - Expanded evaluator generation with optional model selection, dimensions, generated weights, pass thresholds, and
explicit trace IDs. Added preview quality/safety categories, Boolean/Likert rubric scoring, dimension eligibility,
and endpoint evaluator schema-version selection. - Added preview evaluator diagnostics through
BetaEvaluatorsClient.listEvaluatorDiagnosticsand the asynchronous
equivalent, including dimension, definition, version-change, and health diagnostics. - Added preview conversation data generation with sample limits, redaction, descriptions, rubrics, and model options,
plus per-test-case rubric generation for trace jobs. Data-generation results now expose warnings and warning codes. - Added
deleteSkillVersionanddeleteSkillVersionWithResponseto the synchronous and asynchronous beta Skills clients. - Added
JobStatus.SKIPPED.
Breaking Changes
- Replaced
BetaAgentInsightMonitorsClientandBetaAgentInsightMonitorsAsyncClientwith the GA clients. Replace
AIProjectClientBuilder.beta().buildBetaAgentInsightMonitorsClient()and its asynchronous equivalent with
buildAgentInsightMonitorsClient()andbuildAgentInsightMonitorsAsyncClient()on the main builder. - Replaced
AgentInsightMonitorCreateinputs withcreateAgentInsightMonitor(agentName, modelDeploymentName, ...)
overloads. ReplacedAgentInsightMonitorUpdateandAgentInsightUpdatewithUpdateAgentInsightMonitorOptions
andUpdateAgentInsightOptions. - Replaced the
AgentInsightRunCreateconvenience-method input withbeginCreateAgentInsightRun(monitorId, ...)
overloads.AgentInsightRun.getInputs()now returnsAgentInsightRunInputs. - Replaced the positional convenience overloads of
listAgentInsightsandlistAgentInsightRunswith options bags.
Pass the monitor ID in the options object rather than as a separate method argument. - Replaced monitor overview/override models with agent summaries and priorities.
AgentInsightMonitor.getOverview()
is nowgetAgentSummary(), returningAgentInsightSummary; removedAgentInsightsOverview,
AgentInsightsOverviewOverride, andAgentInsightOverviewSource.
RenamedAgentInsightDetails.getRecommendedActions()togetRecommendedAction(). - Renamed cron and recurrence trigger
startTime/endTimeaccessors tostartAt/endAt.
Cron and one-time trigger time zones now usejava.util.TimeZoneinstead ofString. - Replaced
DailyRecurrenceSchedule(List<Integer> hours)with a no-argument constructor andsetHours(...).
Bugs Fixed
- Fixed long-running job polling to recognize the service's
cancelledstatus as terminal.
Other Changes
- Regenerated the client with
@azure-tools/openai-typespecversion1.34.1. - Updated
azure-ai-agentsfrom2.9.0to2.10.0. - Updated
openai-javafrom4.69.0to4.99.0. - Updated
azure-storage-blobfrom12.36.0to12.37.2.