Muokkaa

Jaa

Preview features in Azure AI Search

  • Artikkeli

This article identifies all data plane and control plane features in public preview. This list is helpful for checking feature status. It also explains how to call a preview REST API.

Preview API versions are cumulative and roll up to the next preview. We recommend always using the latest preview APIs for full access to all preview features.

Preview features are removed from this list if they're retired or transition to general availability. For announcements regarding general availability and retirement, see Service Updates or What's New.

Data plane preview features

Feature                         Category Description Availability
Query rewrite in the semantic reranker Relevance (scoring) You can set options on a semantic query to rewrite the query input into a revised or expanded query that generates more relevant results from the L2 ranker. Search Documents (preview).
Document Layout skill Applied AI (skills) A new skill used to analyze a document for structure and provide structure-aware chunking. Create or Update Skillset (preview).
Keyless billing for Azure AI skills processing. Applied AI (skills) You can now use a managed identity and roles for a keyless connection to Azure AI services for built-in skills processing. This capability removes restrictions for having both search and AI services in the same region. Create or Update Skillset (preview).
Markdown parsing mode Indexer data source With this parsing mode, indexers can generate one-to-one or one-to-many search documents from Markdown files in Azure Storage. Create or Update Indexer (preview).
Rescoring options for compressed vectors Relevance (scoring) You can set options to rescore with original vectors instead of compressed vectors. Applies to HNSW and exhaustive KNN vector algorithms, using binary and scalar compression. Create or Update Index (preview).
Lower the dimension requirements for MRL-trained text embedding models on Azure OpenAI Index Text-embedding-3-small and Text-embedding-3-large are trained using Matryoshka Representation Learning (MRL). This allows you to truncate the embedding vectors to fewer dimensions, and adjust the balance between vector index size usage and retrieval quality. A new truncationDimension provides the MRL behaviors as an extra parameter in a vector compression configuration. This can only be configured for new vector fields. Create or Update Index (preview).
Unpack @search.score to view subscores in hybrid search results Relevance (scoring) You can investigate Reciprocal Rank Fusion (RRF) ranked results by viewing the individual query subscores of the final merged and scored result. A new debug property unpacks the search score. QueryResultDocumentSubscores, QueryResultDocumentRerankerInput, and QueryResultDocumentSemanticField provide the extra detail. Search Documents (preview).
Target filters in a hybrid search to just the vector queries Query A filter on a hybrid query involves all subqueries on the request, regardless of type. You can override the global filter to scope the filter to a specific subquery. A new filterOverride parameter provides the behaviors. Search Documents (preview).
Text Split skill (token chunking) Applied AI (skills) This skill has new parameters that improve data chunking for embedding models. A new unit parameter lets you specify token chunking. You can now chunk by token length, setting the length to a value that makes sense for your embedding model. You can also specify the tokenizer and any tokens that shouldn't be split during data chunking. Create or Update Skillset (preview).
Azure AI Vision multimodal embedding skill Applied AI (skills) A new skill type that calls Azure AI Vision multimodal API to generate embeddings for text or images during indexing. Create or Update Skillset (preview).
Azure Machine Learning (AML) skill Applied AI (skills) AML skill integrates an inferencing endpoint from Azure Machine Learning. In previous preview APIs, it supports connections to deployed custom models in an AML workspace. Starting in the 2024-05-01-preview, you can use this skill in workflows that connect to embedding models in the Azure AI Foundry model catalog. It's also available in the Azure portal, in skillset design, assuming Azure AI Search and Azure Machine Learning services are deployed in the same subscription. Create or Update Skillset (preview).
Incremental enrichment cache Applied AI (skills) Adds caching to an enrichment pipeline, allowing you to reuse existing output if a targeted modification, such as an update to a skillset or another object, doesn't change the content. Caching applies only to enriched documents produced by a skillset. Create or Update Indexer (preview).
OneLake files indexer Indexer data source New data source for extracting searchable data and metadata data from a lakehouse on top of OneLake Create or Update Data Source (preview).
Azure Files indexer Indexer data source New data source for indexer-based indexing from Azure Files Create or Update Data Source (preview).
SharePoint Online indexer Indexer data source New data source for indexer-based indexing of SharePoint content. Sign up to enable the feature. Create or Update Data Source (preview) or the Azure portal.
MySQL indexer Indexer data source New data source for indexer-based indexing of Azure MySQL data sources. Sign up to enable the feature. Create or Update Data Source (preview), .NET SDK 11.2.1, and Azure portal.
Azure Cosmos DB for MongoDB indexer Indexer data source New data source for indexer-based indexing through the MongoDB APIs in Azure Cosmos DB. Sign up to enable the feature. Create or Update Data Source (preview) or the Azure portal.
Azure Cosmos DB for Apache Gremlin indexer Indexer data source New data source for indexer-based indexing through the Apache Gremlin APIs in Azure Cosmos DB. Sign up to enable the feature. Create or Update Data Source (preview).
Native blob soft delete Indexer data source Applies to the Azure Blob Storage indexer. Recognizes blobs that are in a soft-deleted state, and removes the corresponding search document during indexing. Create or Update Data Source (preview).
Reset Documents Indexer Reprocesses individually selected search documents in indexer workloads. Reset Documents (preview).
speller Query Optional spelling correction on query term inputs for simple, full, and semantic queries. Search Documents (preview).
Normalizers Query Normalizers provide simple text preprocessing: consistent casing, accent removal, and ASCII folding, without invoking the full text analysis chain. Search Documents (preview).
featuresMode parameter Relevance (scoring) BM25 relevance score expansion to include details: per field similarity score, per field term frequency, and per field number of unique tokens matched. You can consume these data points in custom scoring solutions. Search Documents (preview).
vectorQueries.threshold parameter Relevance (scoring) Exclude low-scoring search result based on a minimum score. Search Documents (preview).
hybridSearch.maxTextRecallSize and countAndFacetMode parameters Relevance (scoring) adjust the inputs to a hybrid query by controlling the amount BM25-ranked results that flow to the hybrid ranking model. Search Documents (preview).
moreLikeThis Query Finds documents that are relevant to a specific document. This feature has been in earlier previews. Search Documents (preview).

Control plane preview features

Feature                         Category Description Availability
Network security perimeter Service Join a search service to a network security perimeter to control network access to your search service. The Azure portal and the Network Security Perimeter APIs 2024-06-01-preview.
Search service under a user-assigned managed identity Service Configures a search service to use a previously created user-assigned managed identity. Services - Update, 2021-04-01-preview, or the latest preview version. We recommend using the latest preview version.

Preview features in Azure SDKs

Each Azure SDK team releases beta packages on their own timeline. Check the change log for mentions of new features in beta packages:

Using preview features

Experimental features are available through the preview REST API first, followed by Azure portal, and then the Azure SDKs.

The following statements apply to preview features:

  • Preview features are available under Supplemental Terms of Use, without a service level agreement.
  • Preview features might undergo breaking changes if a redesign is required.
  • Sometimes preview features don't make it into a GA release.

If you write code against a preview API, you should prepare to upgrade that code to newer API versions when they roll out. We maintain an Upgrade REST APIs document to make that step easier.

How to call a preview REST API

Preview REST APIs are accessed through the api-version parameter on the URI. Older previews are still operational but become stale over time and aren't updated with new features or bug fixes.

For data plane operation on content, 2024-05-01-preview is the most recent preview version. The following example shows the syntax for Indexes GET (preview):

GET {endpoint}/indexes('{indexName}')?api-version=2024-05-01-Preview

For management operations on the search service, 2024-06-01-preview is the most recent preview version. The following example shows the syntax for Update Service 2024-06-01-preview version.

PATCH https://management.azure.com/subscriptions/subid/resourceGroups/rg1/providers/Microsoft.Search/searchServices/mysearchservice?api-version=2024-06-01-preview

{
  "tags": {
    "app-name": "My e-commerce app",
    "new-tag": "Adding a new tag"
  },
  "properties": {
    "replicaCount": 2
  }
}

See also