Azure AI Search client library for Python - version 11.4.0

Azure AI Search (formerly known as "Azure Cognitive Search") is an AI-powered information retrieval platform that helps developers build rich search experiences and generative AI apps that combine large language models with enterprise data.

Azure AI Search is well suited for the following application scenarios:

  • Consolidate varied content types into a single searchable index. To populate an index, you can push JSON documents that contain your content, or if your data is already in Azure, create an indexer to pull in data automatically.
  • Attach skillsets to an indexer to create searchable content from images and unstructured documents. A skillset leverages APIs from Azure AI services for built-in OCR, entity recognition, key phrase extraction, language detection, text translation, and sentiment analysis. You can also add custom skills to integrate external processing of your content during data ingestion.
  • In a search client application, implement query logic and user experiences similar to commercial web search engines.

Use the Azure.Search.Documents client library to:

  • Submit queries using vector, keyword, and hybrid query forms.
  • Implement filtered queries for metadata, geospatial search, faceted navigation, or to narrow results based on filter criteria.
  • Create and manage search indexes.
  • Upload and update documents in the search index.
  • Create and manage indexers that pull data from Azure into an index.
  • Create and manage skillsets that add AI enrichment to data ingestion.
  • Create and manage analyzers for advanced text analysis or multi-lingual content.
  • Optimize results through semantic ranking and scoring profiles to factor in business logic or freshness.

Source code | Package (PyPI) | Package (Conda) | API reference documentation | Product documentation | Samples

Getting started

Install the package

Install the Azure AI Search client library for Python with pip:

pip install azure-search-documents

Prerequisites

To create a new search service, you can use the Azure portal, Azure PowerShell, or the Azure CLI.

az search service create --name <mysearch> --resource-group <mysearch-rg> --sku free --location westus

See choosing a pricing tier for more information about available options.

Authenticate the client

To interact with the search service, you'll need to create an instance of the appropriate client class: SearchClient for searching indexed documents, SearchIndexClient for managing indexes, or SearchIndexerClient for crawling data sources and loading search documents into an index.

To instantiate a client object, you'll need an endpoint and Azure roles or an API key. You can refer to the documentation for more information on supported authenticating approaches with the search service.

Get an API Key

An API key can be an easier approach to start with because it doesn't require pre-existing role assignments.

You can get the endpoint and an API key from the search service in the Azure Portal. Please refer the documentation for instructions on how to get an API key.

Alternatively, you can use the following Azure CLI command to retrieve the API key from the search service:

az search admin-key show --service-name <mysearch> --resource-group <mysearch-rg>

There are two types of keys used to access your search service: admin (read-write) and query (read-only) keys. Restricting access and operations in client apps is essential to safeguarding the search assets on your service. Always use a query key rather than an admin key for any query originating from a client app.

Note: The example Azure CLI snippet above retrieves an admin key so it's easier to get started exploring APIs, but it should be managed carefully.

Create a SearchClient

To instantiate the SearchClient, you'll need the endpoint, API key and index name:

from azure.core.credentials import AzureKeyCredential
from azure.search.documents import SearchClient

service_endpoint = os.environ["AZURE_SEARCH_SERVICE_ENDPOINT"]
index_name = os.environ["AZURE_SEARCH_INDEX_NAME"]
key = os.environ["AZURE_SEARCH_API_KEY"]

search_client = SearchClient(service_endpoint, index_name, AzureKeyCredential(key))

Create a client using Microsoft Entra ID authentication

You can also create a SearchClient, SearchIndexClient, or SearchIndexerClient using Microsoft Entra ID authentication. Your user or service principal must be assigned the "Search Index Data Reader" role. Using the DefaultAzureCredential you can authenticate a service using Managed Identity or a service principal, authenticate as a developer working on an application, and more all without changing code. Please refer the documentation for instructions on how to connect to Azure AI Search using Azure role-based access control (Azure RBAC).

Before you can use the DefaultAzureCredential, or any credential type from Azure.Identity, you'll first need to install the Azure.Identity package.

To use DefaultAzureCredential with a client ID and secret, you'll need to set the AZURE_TENANT_ID, AZURE_CLIENT_ID, and AZURE_CLIENT_SECRET environment variables; alternatively, you can pass those values to the ClientSecretCredential also in Azure.Identity.

Make sure you use the right namespace for DefaultAzureCredential at the top of your source file:

from azure.identity import DefaultAzureCredential
from azure.search.documents import SearchClient

service_endpoint = os.getenv("AZURE_SEARCH_SERVICE_ENDPOINT")
index_name = os.getenv("AZURE_SEARCH_INDEX_NAME")
credential = DefaultAzureCredential()

search_client = SearchClient(service_endpoint, index_name, credential)

Key concepts

An Azure AI Search service contains one or more indexes that provide persistent storage of searchable data in the form of JSON documents. (If you're brand new to search, you can make a very rough analogy between indexes and database tables.) The Azure.Search.Documents client library exposes operations on these resources through two main client types.

Azure AI Search provides two powerful features: semantic ranking and vector search.

Semantic ranking enhances the quality of search results for text-based queries. By enabling semantic ranking on your search service, you can improve the relevance of search results in two ways:

  • It applies secondary ranking to the initial result set, promoting the most semantically relevant results to the top.
  • It extracts and returns captions and answers in the response, which can be displayed on a search page to enhance the user's search experience.

To learn more about semantic ranking, you can refer to the documentation.

Vector search is an information retrieval technique that overcomes the limitations of traditional keyword-based search. Instead of relying solely on lexical analysis and matching individual query terms, vector search utilizes machine learning models to capture the contextual meaning of words and phrases. It represents documents and queries as vectors in a high-dimensional space called an embedding. By understanding the intent behind the query, vector search can deliver more relevant results that align with the user's requirements, even if the exact terms are not present in the document. Moreover, vector search can be applied to various types of content, including images and videos, not just text.

To learn how to index vector fields and perform vector search, you can refer to the sample. This sample provides detailed guidance on indexing vector fields and demonstrates how to perform vector search.

Additionally, for more comprehensive information about Vector Search, including its concepts and usage, you can refer to the documentation. The documentation provides in-depth explanations and guidance on leveraging the power of Vector Search in Azure AI Search.

The Azure.Search.Documents client library (v1) is a brand new offering for Python developers who want to use search technology in their applications. There is an older, fully featured Microsoft.Azure.Search client library (v10) with many similar looking APIs, so please be careful to avoid confusion when exploring online resources.

Examples

The following examples all use a simple Hotel data set that you can import into your own index from the Azure portal. These are just a few of the basics - please check out our Samples for much more.

Querying

Let's start by importing our namespaces.

import os
from azure.core.credentials import AzureKeyCredential
from azure.search.documents import SearchClient

We'll then create a SearchClient to access our hotels search index.

index_name = "hotels"
# Get the service endpoint and API key from the environment
endpoint = os.environ["SEARCH_ENDPOINT"]
key = os.environ["SEARCH_API_KEY"]

# Create a client
credential = AzureKeyCredential(key)
client = SearchClient(endpoint=endpoint,
                      index_name=index_name,
                      credential=credential)

Let's search for a "luxury" hotel.

results = client.search(search_text="luxury")

for result in results:
    print("{}: {})".format(result["hotelId"], result["hotelName"]))

Creating an index

You can use the SearchIndexClient to create a search index. Fields can be defined using convenient SimpleField, SearchableField, or ComplexField models. Indexes can also define suggesters, lexical analyzers, and more.

client = SearchIndexClient(service_endpoint, AzureKeyCredential(key))
name = "hotels"
fields = [
    SimpleField(name="hotelId", type=SearchFieldDataType.String, key=True),
    SimpleField(name="baseRate", type=SearchFieldDataType.Double),
    SearchableField(name="description", type=SearchFieldDataType.String, collection=True),
    ComplexField(
        name="address",
        fields=[
            SimpleField(name="streetAddress", type=SearchFieldDataType.String),
            SimpleField(name="city", type=SearchFieldDataType.String),
        ],
        collection=True,
    ),
]
cors_options = CorsOptions(allowed_origins=["*"], max_age_in_seconds=60)
scoring_profiles: List[ScoringProfile] = []
index = SearchIndex(name=name, fields=fields, scoring_profiles=scoring_profiles, cors_options=cors_options)

result = client.create_index(index)

Adding documents to your index

You can Upload, Merge, MergeOrUpload, and Delete multiple documents from an index in a single batched request. There are a few special rules for merging to be aware of.

DOCUMENT = {
    "category": "Hotel",
    "hotelId": "1000",
    "rating": 4.0,
    "rooms": [],
    "hotelName": "Azure Inn",
}

result = search_client.upload_documents(documents=[DOCUMENT])

print("Upload of new document succeeded: {}".format(result[0].succeeded))

Authenticate in a National Cloud

To authenticate in a National Cloud, you will need to make the following additions to your client configuration:

  • Set the AuthorityHost in the credential options or via the AZURE_AUTHORITY_HOST environment variable
  • Set the audience in SearchClient, SearchIndexClient, or SearchIndexerClient
# Create a SearchClient that will authenticate through AAD in the China national cloud.
import os
from azure.identity import DefaultAzureCredential, AzureAuthorityHosts
from azure.search.documents import SearchClient

index_name = "hotels"
endpoint = os.environ["SEARCH_ENDPOINT"]
key = os.environ["SEARCH_API_KEY"]
credential = DefaultAzureCredential(authority=AzureAuthorityHosts.AZURE_CHINA)

search_client = SearchClient(endpoint, index_name, credential=credential, audience="https://search.azure.cn")

Retrieving a specific document from your index

In addition to querying for documents using keywords and optional filters, you can retrieve a specific document from your index if you already know the key. You could get the key from a query, for example, and want to show more information about it or navigate your customer to that document.

from azure.core.credentials import AzureKeyCredential
from azure.search.documents import SearchClient

search_client = SearchClient(service_endpoint, index_name, AzureKeyCredential(key))

result = search_client.get_document(key="23")

print("Details for hotel '23' are:")
print("        Name: {}".format(result["hotelName"]))
print("      Rating: {}".format(result["rating"]))
print("    Category: {}".format(result["category"]))

Async APIs

This library includes a complete async API. To use it, you must first install an async transport, such as aiohttp. See azure-core documentation for more information.

from azure.core.credentials import AzureKeyCredential
from azure.search.documents.aio import SearchClient

search_client = SearchClient(service_endpoint, index_name, AzureKeyCredential(key))

async with search_client:
    results = await search_client.search(search_text="spa")

    print("Hotels containing 'spa' in the name (or other fields):")
    async for result in results:
        print("    Name: {} (rating {})".format(result["hotelName"], result["rating"]))

Troubleshooting

General

The Azure AI Search client will raise exceptions defined in Azure Core.

Logging

This library uses the standard logging library for logging. Basic information about HTTP sessions (URLs, headers, etc.) is logged at INFO level.

Detailed DEBUG level logging, including request/response bodies and unredacted headers, can be enabled on a client with the logging_enable keyword argument:

import sys
import logging
from azure.core.credentials import AzureKeyCredential
from azure.search.documents import SearchClient

# Create a logger for the 'azure' SDK
logger = logging.getLogger('azure')
logger.setLevel(logging.DEBUG)

# Configure a console output
handler = logging.StreamHandler(stream=sys.stdout)
logger.addHandler(handler)

# This client will log detailed information about its HTTP sessions, at DEBUG level
client = SearchClient("<service endpoint>", "<index_name>", AzureKeyCredential("<api key>"), logging_enable=True)

Similarly, logging_enable can enable detailed logging for a single operation, even when it isn't enabled for the client:

result =  client.search(search_text="spa", logging_enable=True)

Next steps

Contributing

See our Search CONTRIBUTING.md for details on building, testing, and contributing to this library.

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit cla.microsoft.com.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Impressions

Impressions