The AI Feature That Almost Every Database Missed
While every database vendor rushed to VECTORIZE their products, there's another key feature that seems to be ignored - one that's at least as important as vector capabilities for AI. I did some quick research and found that only Snowflake supports it natively:
CREATE TAG data_sensitivity;
ALTER TABLE customer_features SET TAG data_sensitivity = 'high';
It's the ability to attach arbitrary metadata to database objects in a structured way.
This capability would allow organizations to store annotations directly alongside the database objects they describe. Such contextual information is increasingly critical as AI systems become more integrated with enterprise data infrastructure, requiring rich metadata to ensure proper governance, explainability, and compliance. To phrase it in AI-speak, it's all about providing CONTEXT.
Let's look at the big four databases according to the DB-Engines ranking:
Oracle: Oracle Database offers limited metadata support through its Data Dictionary and user-defined properties in Oracle Multimedia. However, it lacks a comprehensive, flexible system for attaching arbitrary metadata to standard database objects like tables, columns, and views.
SQL Server: Microsoft SQL Server provides extended properties through the sp_addextendedproperty stored procedure. This comes closest among traditional RDBMSs to offering metadata attachment.
MySQL: Nothing here. Nada.
PostgreSQL: SECURITY LABEL is the closest thing, but it wasn't built for this purpose.
The common hack is to use COMMENT
. The problem with COMMENT
is it's just a plain text field that's very easy to mess up if developers, DBAs, security teams, and data engineers all want to add their favorite metadata.
Part of the blame falls on the SQL Standard, which doesn't define this capability. Snowflake probably supports it because classifying database objects is critical for data warehouse use cases.
Providing context to LLMs is becoming a primary use case for databases. I wish Snowflake's CREATE TAG
-like syntax would be incorporated into future database systems.
After all, in the age of AI, context isn't just king - it's the whole kingdom.