Preview features in Azure AI Search

2025-09-28

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
"Fast path" for knowledge agents	Agentic search	New `attemptFastPath` boolean for knowledge agents. Enables a shorter processing time if queries are concise and the initial response is sufficiently relevant.	Knowledge Agents - Create Or Update (preview)
Retrieval instructions	Agentic search	New `retrievalInstructions` property for knowledge agents guides query planning in an agentic retrieval workflow. For example, you can specify criteria for including or excluding specific knowledge sources.	Knowledge Agents - Create Or Update (preview)
Improved indexer runtime tracking information	Indexers	New cumulative indexer processing information for the search service and for specific indexers.	Get Service Statistics (preview) and Get Status - Indexers (preview)
Strict postfiltering for vector queries	Vectors	New `strictPostFilter` mode for the `vectorFilterMode` parameter. When specified, filters are applied after the global top-`k` vector results are identified, ensuring that returned documents are a subset of the unfiltered results.	Search Documents (preview).
Agentic retrieval	Query	Create a conversational search experience powered by large language models (LLMs) and your proprietary content. Agentic retrieval breaks down complex user queries into subqueries, runs the subqueries simultaneously, and either extracts grounding data or synthesizes an answer based on documents indexed in Azure AI Search. To get started, see Quickstart: Agentic retrieval. The pipeline involves one or more knowledge sources and an associated knowledge agent, whose response payload provides full transparency into the query plan and reference data. Knowledge sources currently support search indexes and Azure blobs.	Knowledge Sources (preview), Knowledge Agents (preview), and Knowledge Retrieval (preview).
Multivector support	Indexing	Index multiple child vectors within a single document field. You can now use vector types in nested fields of complex collections, effectively allowing multiple vectors to be associated with a single document.	Create or Update Index (preview).
Document-level access control	Security	Flow document-level permissions from blobs in Azure Data Lake Storage (ADLS) Gen2 to searchable documents in an index. Queries can now filter results based on user identity for selected data sources.	Create or Update Index (preview).
GenAI Prompt skill	Skills	A new skill that connects to a large language model (LLM) for information, using a prompt you provide. With this skill, you can populate a searchable field using content from an LLM. A primary use case for this skill is image verbalization, using an LLM to describe images and send the description to a searchable field in your index.	Create or Update Skillset (preview).
flightingOptIn parameter in a semantic configuration	Queries	You can opt in to use prerelease semantic ranking models if one is available in a search service region.	Create or Update Index (preview).
Facet hierarchies, aggregations, and facet filters	Queries	New facet query parameters support nested facets. For numeric facetable fields, you can sum the values of each field. You can also specify filters on a facet to add inclusion or exclusion criteria.	Search Documents (preview).
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).
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).
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).
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).
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

Currently, there are no control plane features in preview.

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 operations on content, 2025-08-01-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, 2025-05-01-preview is the most recent preview version. The following example shows the syntax for Update Service 2025-05-01-preview version.

PATCH https://management.azure.com/subscriptions/subid/resourceGroups/rg1/providers/Microsoft.Search/searchServices/mysearchservice?api-version=2025-05-01-preview

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

Feedback

Was this page helpful?