LlamaIndex Vector Store: Azure AI Search

0.5.0 verified Mon Apr 27 auth: no python

Integration between LlamaIndex and Azure AI Search (formerly Azure Cognitive Search) for vector storage and hybrid search. Current version: 0.5.0. Released as part of the LlamaIndex plugin ecosystem; updates generally follow the LlamaIndex core release cadence.

pip install llama-index-vector-stores-azureaisearch

Common errors

error ImportError: cannot import name 'AzureAISearchVectorStore' from 'llama_index.vector_stores' ↓

cause Wrong import path: the class is in a submodule, not directly in vector_stores.

fix

Use 'from llama_index.vector_stores.azureaisearch import AzureAISearchVectorStore'.

error azure.core.exceptions.ServiceRequestError: ... The request failed due to connection failure ↓

cause Azure AI Search endpoint is unreachable or credentials are missing/incorrect.

fix

Verify AZURE_SEARCH_ENDPOINT environment variable and that DefaultAzureCredential is configured.

error azure.core.exceptions.HttpResponseError: (InvalidRequest) ... The request is invalid. ↓

cause Index schema mismatch (e.g., missing vector configuration or field type).

fix

Check that the Azure AI Search index has a vector field with correct dimensionality and HNSW configuration.

Warnings

breaking Breaking change in 0.4.0: `index_name` is no longer inferred from other parameters; must be explicitly provided. ↓

fix Always pass `index_name` when constructing AzureAISearchVectorStore.

deprecated The `search_or_index_endpoint` parameter is deprecated in favor of `search_service_endpoint` and `index_name` combination (some versions). Check current docs. ↓

fix Use `search_service_endpoint` and `index_name` individually if required.

gotcha DefaultAzureCredential fails in non-Azure environments (local dev) unless environment variables AZURE_TENANT_ID, AZURE_CLIENT_ID, AZURE_CLIENT_SECRET are set. ↓

fix Use AzureCliCredential or EnvironmentCredential with explicit environment variables for local testing.

gotcha Embedding dimensionality must match the Azure AI Search index field definition. Mismatch causes silent query errors. ↓

fix Ensure `embedding_dimensionality` matches the index schema (e.g., 1536 for text-embedding-ada-002).

Imports

AzureAISearchVectorStore
wrong
```
from llama_index.vector_stores import AzureAISearchVectorStore
```
correct
```
from llama_index.vector_stores.azureaisearch import AzureAISearchVectorStore
```
The vector store class is nested under the azureaisearch submodule; direct import from vector_stores package fails.

AzureAISearchIndex

from llama_index.vector_stores.azureaisearch import AzureAISearchIndex

Correct import path; no common wrong pattern known.

Quickstart

Minimal setup to create an AzureAISearchVectorStore using DefaultAzureCredential.

import os
from llama_index.vector_stores.azureaisearch import AzureAISearchVectorStore
from azure.identity import DefaultAzureCredential

# Use environment variables for credentials
endpoint = os.environ.get('AZURE_SEARCH_ENDPOINT', 'https://mysearch.search.windows.net')
credential = DefaultAzureCredential()

vector_store = AzureAISearchVectorStore(
    search_or_index_endpoint=endpoint,
    credential=credential,
    index_name='my-index',
    embedding_dimensionality=1536
)
# Then integrate with LlamaIndex StorageContext