Skip to main content

Azure AI Search

https://azure.microsoft.com/en-us/products/ai-services/ai-search/

Maven Dependency​

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-azure-ai-search</artifactId>
    <version>1.20.0-beta30</version>
</dependency>

APIs​

  • AzureAiSearchEmbeddingStore - supports vector search
  • AzureAiSearchContentRetriever - supports vector, full-text, hybrid searches and re-ranking

Metadata from a bring-your-own index​

By default AzureAiSearchContentRetriever and AzureAiSearchEmbeddingStore read segment metadata from the nested metadata field they write themselves. When you point them at a pre-existing index (createOrUpdateIndex(false)) whose fields are stored at the document root, use metadataFieldNames on either builder to list the top-level fields to expose as metadata:

ContentRetriever retriever = AzureAiSearchContentRetriever.builder()
        .endpoint(endpoint)
        .tokenCredential(tokenCredential)
        .indexName(indexName)
        .createOrUpdateIndex(false)
        .queryType(AzureAiSearchQueryType.FULL_TEXT)
        .metadataFieldNames(List.of("sourcepage", "weburl", "topic", "role"))
        .maxResults(4)
        .build();

Each listed field is copied into the segment Metadata when its value is a type Metadata supports (String, Integer, Long, Float, Double, or UUID). Fields that are absent, null or of another type (for example the embedding vector) are skipped. A field must be marked as retrievable in the Azure index for its value to be returned, and when a field name matches a key in the nested metadata, the top-level value takes precedence.

This option only configures metadata extraction; the existing requirements for the index's core ID, content, vector, and semantic-search fields remain unchanged.

Examples​