nlweb-azure-vectordb

Azure AI Search provider for NLWeb.

Overview

This is a blueprint package demonstrating how to create individual provider packages for NLWeb. It contains only the Azure AI Search retrieval provider.

Third-party developers can use this as a template for creating their own provider packages (e.g., nlweb-pinecone, nlweb-weaviate, etc.).

Installation

pip install nlweb-core nlweb-azure-vectordb

For LLM and embedding, you'll also need a model provider:

pip install nlweb-azure-models

Or use the bundle packages:

pip install nlweb-core nlweb-retrieval nlweb-models

Configuration

Create config.yaml:

retrieval:
  provider: azure_ai_search
  import_path: nlweb_azure_vectordb.azure_search_client
  class_name: AzureSearchClient
  api_endpoint_env: AZURE_SEARCH_ENDPOINT
  auth_method: azure_ad  # or api_key
  index_name: my-search-index

Authentication Methods

API Key Authentication

retrieval:
  provider: azure_ai_search
  import_path: nlweb_azure_vectordb.azure_search_client
  class_name: AzureSearchClient
  api_endpoint_env: AZURE_SEARCH_ENDPOINT
  api_key_env: AZURE_SEARCH_KEY
  auth_method: api_key
  index_name: my-index

Set environment variables:

export AZURE_SEARCH_ENDPOINT=https://your-service.search.windows.net
export AZURE_SEARCH_KEY=your_key_here

Managed Identity (Azure AD) Authentication

retrieval:
  provider: azure_ai_search
  import_path: nlweb_azure_vectordb.azure_search_client
  class_name: AzureSearchClient
  api_endpoint_env: AZURE_SEARCH_ENDPOINT
  auth_method: azure_ad
  index_name: my-index

Set environment variable:

export AZURE_SEARCH_ENDPOINT=https://your-service.search.windows.net

Usage

import nlweb_core

# Initialize
nlweb_core.init(config_path="./config.yaml")

# Search
from nlweb_core import retriever

results = await retriever.search(
    query="example query",
    site="example.com",
    num_results=10
)

Features

• Vector similarity search with Azure AI Search
• Hybrid search (vector + keyword)
• Managed identity (Azure AD) authentication support
• API key authentication support
• Configurable index names
• Compatible with NLWeb Protocol v0.5

Creating Your Own Provider Package

Use this package as a template:

1. Create package structure:

   nlweb-yourprovider/
   ├── pyproject.toml
   ├── README.md
   └── nlweb_yourprovider/
       ├── __init__.py
       └── your_client.py
   

2. Implement VectorDBClientInterface:

   from nlweb_core.retriever import VectorDBClientInterface
   
   class YourClient(VectorDBClientInterface):
       async def search(self, query, site, num_results, **kwargs):
           # Your implementation
           pass
   

3. Declare dependencies in pyproject.toml:

   dependencies = [
       "nlweb-core>=0.5.0",
       "your-provider-sdk>=1.0.0",
   ]
   

4. Publish to PyPI:

   python -m build
   twine upload dist/*
   

License

MIT License - Copyright (c) 2025 Microsoft Corporation