12.2.0b1 (2026-11-14)
Features Added
- Added
ApiVersion.V2026_11_01_PREVIEWand made2026-11-01-previewthe default.
Previously advertised API versions remain selectable. - Restored explicit
ApiVersion.V2026_10_01selection without changing the November default. - Added
indexes.types.KnowledgeBaseInputandKnowledgeBaseModelInputfor all
three model variants in dictionary-style knowledge-base inputs, while preserving
the callable GAindexes.types.KnowledgeBaseModelalias and the legacy
indexes.types.KnowledgeBase.modelslist type. - Below operations and models are added for graph querying, index configuration, and retrieval:
SearchClient.graph(synchronous and asynchronous), including typed provenance,
diagnostics, evidence, statistics, and explain results.indexes.models.GraphFieldMappingsandindexes.models.GraphProfile.indexes.models.KnowledgeSourceGraphExtractionParameters, grounding, and ontology models.knowledgebases.models.AgentRuntimeGraphParams, graph hydration and provenance parameters.knowledgebases.models.KnowledgeBaseGraphSearchActivityRecordand
knowledgebases.models.KnowledgeBaseGraphSearchReference.
- Below models and properties are added for retrieval and query configuration:
indexes.models.KnowledgeBaseMicrosoftRetriever1Modeland
indexes.models.KnowledgeBaseOpenAICompatibleModel.indexes.models.KnowledgeBaseTooland knowledge-source tool references.indexes.models.OpenAICompatibleVectorizer.indexes.models.SearchIndexKnowledgeSourceFilterHint.auto_configurationand
SearchIndexKnowledgeSourceFilterHintAutoConfigurationfor facet-grounded hints.indexes.models.VectorSearchTopicGraphProfileand none, diverse, and balanced
vector-query topic-graph models indocuments.models.knowledgebases.models.KnowledgeSourceCacheand
knowledgebases.models.KnowledgeSourceOpenAICompatibleVectorizer.models.AgenticSearchOptions, agentic debug information, andQueryType.AGENTIC.
- Below operations and models are added for asynchronous File knowledge-source ingestion:
SearchIndexClient.begin_ingest_knowledge_source_file_multipartand
SearchIndexClient.begin_ingest_knowledge_source_file_replacement, including async clients.- File ingestion operation status/listing, capacity, and error models.
- Added
indexes.models.RandomScoringFunctionandRandomScoringParameters. - Added
indexes.types.ScoringFunctionInput,ScoringProfileInput, andSearchIndexInput
for random scoring in dictionary-style inputs, while preserving the classic
ScoringFunctionalias and legacy scoring-profile and index field types. - Added
more_like_thisandagentic_optionsto synchronous and asynchronous
SearchClient.search. Restored preview query, skill, knowledge-source, and indexer
surfaces remain present in the generated package.
Breaking Changes
These changes do not impact the API of stable versions such as 12.1.0.
Only code written against a beta version such as 12.1.0b2 may be affected.
- The service removes stored
query_hintsand request-timequery_hint_overrides
from indexed knowledge sources other than search-index knowledge sources. - The service removes
query_hint_processingfrom knowledge-base activity models
other than search-index activity models. KnowledgeBaseRetrievalClient.retrieve_streamand its asynchronous equivalent now
yield typed payloads directly instead ofKnowledgeBaseRetrievalEventenvelopes,
matching the October GA contract.
Bugs Fixed
- Removed obsolete retrieval TypedDict annotation overrides that advertised unsupported
queryHintOverridesfields and prevented resolution of the generated search-index query-hint type. - Serialized asynchronous buffered uploads and waited for active uploads before
closing the client. - Closed typed retrieval streams when a response callback raises, including
awaiting asynchronous response cleanup. - Corrected graph query samples to pass request identifiers and content negotiation
through HTTP headers. - Preserved existing resource
get_*list wrappers after generated method renames. - Preserved
ScoringFunction.field_nameandboostcompatibility without adding
those fields to random scoring. - Corrected generated File-ingestion poller namespace references through the
package's reproducible post-generation workaround. - Preserved all buffered indexing actions after request failures and prevented
asynchronous auto-flush from cancelling its own task. - Kept batch wire values synchronized and accepted dictionary batches, including
custom error mappings during oversized-batch splitting. - Preserved the first search response's metadata and continuation across page iteration.
- Corrected semantic answer and caption enum encoding and threshold separators.
- Forwarded conditional-delete options for resource names and extra field-helper properties.
- Preserved buffered client audience, API version, transport options, and schema-client ownership.
- Restored registered GA model-name aliases, TypedDict and sequence inputs, constructor overloads,
and the callableindexes.types.KnowledgeBaseModelalias.