Azure Search Documents Py

Install

Quick Start

Non-Obvious Patterns

• SearchIndexingBufferedSender for batch uploads with auto-batching/retries
• Vector field type: Collection(Edm.Single) with vector_search_dimensions + vector_search_profile_name
• Async client: from azure.search.documents.aio import SearchClient
• KnowledgeBaseRetrievalClient for agentic retrieval with LLM-powered Q&A

Best Practices

1. Use hybrid search for best relevance combining vector and keyword
2. Enable semantic ranking for natural language queries
3. Index in batches of 100-1000 documents for efficiency
4. Use filters to narrow results before ranking
5. Configure vector dimensions to match your embedding model
6. Use HNSW algorithm for large-scale vector search
7. Create suggesters at index creation time (cannot add later)
8. Use SearchIndexingBufferedSender for batch uploads
9. Always define semantic configuration for agentic retrieval indexes
10. Use create_or_update_index for idempotent index creation
11. Close clients with context managers or explicit close()