SQL Server - Flexible AI Model Management

In an era where artificial intelligence is reshaping enterprise operations, Microsoft has positioned SQL Server 2025 as a cornerstone of its AI strategy by introducing Flexible AI Model Management—a feature that fundamentally reimagines how databases interact with machine learning ecosystems. Flexible AI Model Management is a pivotal feature in SQL Server 2025, designed to work seamlessly with Native Vector Support. It revolutionizes the way you integrate and operationalize machine learning models directly within the database. This document outlines the capabilities and benefits of this new feature.

Overview

Flexible AI Model Management offers a unified framework for registering, managing, and invoking AI models directly from within SQL Server, irrespective of their hosting location. This capability streamlines the integration of AI into database workflows, enhancing efficiency and reducing complexity.

Key Capabilities - Model Registration

The feature allows you to register AI models within SQL Server's metadata. This registration process involves specifying the model's location, type, and any necessary metadata. The model itself can reside in various locations, such as:

• External Model Stores: Models hosted in external services like Azure Machine Learning, Amazon SageMaker, or Google AI Platform.

• Local File System: Models stored on the SQL Server's file system.

• Azure Blob Storage: Models stored in Azure Blob Storage.

Key Capabilities - Model Management

Once registered, models can be managed directly within SQL Server. This includes:

• Versioning: Tracking different versions of a model.

• Metadata Management: Storing and updating model metadata, such as descriptions, input/output schemas, and performance metrics.

• Access Control: Managing permissions to control who can access and use the models.

Key Capabilities - Model Invocation

Registered models can be invoked directly from T-SQL using new functions and stored procedures. This allows you to seamlessly integrate AI models into your database queries and applications. The invocation process handles:

• Data Transformation: Automatically transforming data from SQL Server into the format expected by the model.

• Model Execution: Executing the model and retrieving the results.

• Result Transformation: Transforming the model's output back into a SQL Server-compatible format.