12.1.0 (2026-11-13)
Features Added
- Added support for the
2026-10-01service version, which is now the default. - Added
SearchOptions.MoreLikeThisto search for documents similar to an existing document. - Added
querySourceAuthorizationandenableElevatedReadoverloads for document search,GetDocumentOptions.QuerySourceAuthorizationandEnableElevatedReadfor document retrieval, andquerySourceAuthorizationon knowledge base retrieval. - Added document-level permission configuration through
SearchField.PermissionFilter,OrganizationAccessExpiration, andSharepointSiteUrl;SearchIndex.PermissionFilterOption; andSharePointConnectorAppRegistration. Permission filters can also be configured throughSimpleFieldandSimpleFieldAttribute. - Added
SearchIndexerDataSourceConnection.IndexerPermissionOptions,IndexingParametersConfiguration.RefreshAllAcls, andSearchIndexerDataUserAssignedIdentity.FederatedIdentityClientIdfor permission ingestion and federated identity configuration. - Added
SearchIndexerClient.ResyncandResyncAsync, withIndexerResyncBodyandIndexerResyncOption.Permissions, to resynchronize indexed permissions. - Added
SearchIndexerStatus.RuntimeandSearchServiceStatistics.IndexersRuntime, exposing used and remaining runtime and the associated reporting period. - Added
GetAllIndexers,GetAllDataSourceConnections,GetAllSkillsets, andGetAllSynonymMaps, with async counterparts, to expose pageable resource listings with property selection, prefix search, and page-size controls without changing the existing list-returning APIs. - Added
GetAllIndexStatsSummaryandGetAllKnowledgeSourceFiles, with async counterparts, for pageable index statistics and knowledge source file listings. - Added
search,pageSize, andsearchTypecontrols to the existing pageableGetIndexes,GetIndexesWithSelectedProperties,GetAliases,GetKnowledgeBases, andGetKnowledgeSourcesmethods, with async counterparts.ListingSearchType.Prefixsupports filtering listings by name prefix. - Added file knowledge sources through
FileKnowledgeSource,FileKnowledgeSourceParameters, andFileKnowledgeSourceParams, including ingestion and CORS configuration. - Added multipart file upload, update, and deletion through
UploadKnowledgeSourceFileMultipart,UpdateKnowledgeSourceFile, andDeleteKnowledgeSourceFile, with async counterparts.FileUploadMetadataandKnowledgeSourceFileexpose file metadata, prefixes, parsing and extraction modes, timestamps, and processing errors. Multipart request constructors and content properties are experimental (SCME0004). - Added
KnowledgeSourceFileCapacityandKnowledgeSourceStatus.FileCapacityfor maximum file size, maximum file count, and remaining file count. - Added indexed SQL and indexed SharePoint knowledge sources, their retrieval parameter and reference models, SQL content and embedding column mappings, and SharePoint container selection.
- Added
KnowledgeSourceIngestionParameters.IngestionPermissionOptionsandNetworkAccessMode, andSearchIndexKnowledgeSourceParameters.BaseFilter, to configure knowledge source access and filtering. - Added low and medium retrieval reasoning effort, answer-synthesis and extractive-data output modes, retrieval and answer instructions, and CORS configuration on knowledge bases. Retrieval requests now support messages, reasoning effort, output mode, and maximum output document counts.
- Added
AlwaysQuerySource,FailOnError, andMaxOutputDocumentscontrols to knowledge source retrieval parameters, andLanguage,Market,Count, andFreshnessto web knowledge source configuration. - Added typed knowledge base retrieval activity records for search index, Azure Blob, indexed OneLake, web, file, indexed SharePoint, and indexed SQL sources; model query-planning and answer-synthesis activity records; activity timestamps and warnings; model and deployment details; and citation URLs on supported references.
- Added
KnowledgeBaseRetrievalClient.RetrieveStreamAsyncfor typed or protocol server-sent event streaming, with retrieval, activity, answer, reference, response-completion, error, and unknown-event models sharingKnowledgeBaseRetrievalStreamEvent. - Added
ContentUnderstandingSkill.ModelName,ModelDeployment, and chunking method selection throughContentUnderstandingSkillChunkingProperties.MethodandContentUnderstandingSkillChunkingMethod, including fixed-size and semantic chunking. Chunking units now include tokens. - Added
Gpt54,Gpt55,Gpt56Sol,Gpt56Terra, andGpt56LunatoAzureOpenAIModelName.
Bugs Fixed
- Fixed ambiguous API key and configuration-based client registration extension calls while retaining the legacy static registration methods.
- Restored the 12.0.0 optional arguments on the
VectorQuery,KnowledgeSource,FacetResult, and enumerable-basedSearchIndexmodel factories. - Restored the 12.0.0 optional arguments on
SearchModelFactory.KnowledgeBaseAgenticReasoningActivityRecord. The overload accepting activity timestamps and a warning requires all arguments to avoid ambiguity with existing calls. - Fixed explicit
AddSearchClient,AddSearchIndexClient, andAddSearchIndexerClientregistrations withTokenCredential, includingDefaultAzureCredential. (#63072) - Resource listing methods now honor
AsPages(pageSizeHint)on initial and continuation requests while preserving the service-provided cursor. - Fixed indexing configuration round trips, including
RefreshAllAcls, markdown settings, custom dictionary entries, and the legacyIndexingParameters.Configurationadapter on deserialized models. - Fixed semantic and vector search options being lost during deserialization and continuation-token round trips.
- Fixed query-answer thresholds using culture-dependent formatting and parsing, and value facets failing to convert compatible numeric types.
- Fixed settings-based clients ignoring custom document serializers and the
QueryAnswerResultmodel factory rejecting read-only-only dictionaries.
Other Changes
AzureMachineLearningVectorizer,AzureMachineLearningParameters,AIFoundryModelCatalogName,VectorizableImageUrlQuery,VectorizableImageBinaryQuery,KnowledgeBaseAgenticReasoningActivityRecord.RetrievalReasoningEffort, and theirSearchModelFactorymethods are now obsolete. They shipped in 12.0.0 because of an error in the service specification and are not supported by any service version this library targets.- The obsolete AML vectorizer and image query models retain request serialization and typed deserialization, but deserialize as unknown models through their polymorphic base types. The obsolete activity record's
RetrievalReasoningEffortproperty is retained for mocking and is not populated from JSON. - Added experimental (
SCME0002) generic-host registration extensions, including keyed registration, forSearchClient,SearchIndexClient,SearchIndexerClient, andKnowledgeBaseRetrievalClient. - Added constructors and setters that allow knowledge base message roles, text content, image content, and image URLs to be updated.