databricks-ai-search

What it is and what it does

This is the official Python SDK for Databricks AI Search, the managed vector and keyword search service within the Databricks platform. It provides a client to create search endpoints, define delta-sync indexes backed by Databricks tables, and run similarity searches against stored embeddings. The package handles authentication automatically in Databricks notebooks and supports explicit credential methods (PAT, service principals, OAuth) for external use.

The SDK is a rename and continuation of the legacy `databricks-vectorsearch` package; the old class names are preserved as aliases for backward compatibility, but new projects should use the `databricks-ai-search` package directly. It depends on mlflow-skinny, requests, deprecation, and httpx, all lightweight and stable libraries.

Use it for:

• Build semantic search over Databricks Delta tables by creating vector indexes and querying them with embedding vectors.
• Integrate vector search into Databricks notebooks for retrieval-augmented generation (RAG) workflows.
• Manage multiple search endpoints and indexes programmatically from Python without leaving the Databricks platform.
• Query both vector and keyword search from external applications using PAT or service principal authentication.
• Migrate from the legacy `databricks-vectorsearch` package to the new `databricks-ai-search` naming without code changes.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Python client for Databricks AI Search, a managed vector and keyword search service on the Databricks platform, supporting endpoint creation, index management, and similarity search queries.

Yes, if you are already using Databricks and need vector search. The package is actively maintained, has low install friction, and integrates seamlessly with Databricks notebooks. However, the license is restrictive and proprietary—use is only permitted in connection with a Databricks Platform Services agreement. Do not install if you are not a Databricks customer or do not have a valid service agreement in place.

Install

databricks-ai-search on PyPI

pip install databricks-ai-search

uv add databricks-ai-search

poetry add databricks-ai-search

Installing databricks-ai-search

Before you install

Low install friction; depends on four lightweight runtime packages (mlflow-skinny, requests, deprecation, httpx). Actively maintained with recent releases.

License in practice

License is unclear and proprietary: use is restricted to connection with Databricks Platform Services under a separate agreement. You cannot use this package standalone or outside that contractual relationship.

Quickstart

pip install databricks-ai-search

from databricks.ai_search.client import AISearchClient

client = AISearchClient()
results = client.create_endpoint(name="my-endpoint", endpoint_type="STANDARD")

Requires a Databricks workspace and valid authentication (auto-detected in notebooks, or explicit credentials via workspace_url and personal_access_token or service principal parameters).

Verify before relying

• Whether the package works outside Databricks notebooks or requires a Databricks workspace
• Specific performance characteristics or rate limits for similarity search queries
• Whether backward-compat aliases (VectorSearchClient, etc.) are fully equivalent or have subtle differences

Package facts

Evidence: databricks_ai_search-0.78-py3-none-any.whl

Tags

More Database packages