The ai_search function can now search Lakebase synced tables

The ai_search function can now retrieve data directly from Lakebase synced tables that have Lakebase Search enabled, expanding available knowledge sources.

Use cases

Imagine you need to enrich AI responses using operational relational data synced from the lakehouse. With this update, you can:

• Combine searches across Lakebase synced tables and vector search indices in the same call.
• Query low-latency operational data directly through ai_search.

How to use

Add a lakebase_search knowledge source in your ai_search function call to query the synced table.

SELECT ai_search(
  'my_question_column',
  array(
    named_struct('type', 'lakebase_search', 'name', 'my_catalog.my_schema.my_synced_table')
  )
);

What happens in this code:

1. The ai_search function receives the question text and the list of knowledge sources.
2. The lakebase_search parameter points to the synced table in Unity Catalog.
3. The search runs against the Lakebase data to return relevant context.

Limitations and requirements

• This feature is in Beta.
• Requires using the Databricks ai_search function.