databricks Databricks Release Notes ·

Databricks ai_prep_search Function Now Accepts Plain Text and Markdown (Beta)

aidatabrickspreviewdata-scientist
feature

The Databricks `ai_prep_search()` SQL function now accepts plain text and markdown string content, expanding beyond its previous requirement for structured VARIANT output. This enhancement allows developers to directly chunk and prepare existing textual data for AI Search indexing and retrieval-augmented generation (RAG) pipelines. It streamlines the process by removing the prerequisite of document parsing, enabling more flexible data preparation workflows. This capability is currently available in Beta.

Features (1)
  • ai_prep_search supports plain text and markdown input

    The `ai_prep_search()` SQL function has been updated to accept plain text and markdown string content, removing the previous dependency on `ai_parse_document`'s structured output. This allows direct preparation of existing text for AI Search indexing and RAG pipelines without prior parsing.

Read the original announcement →

https://docs.databricks.com/aws/en/release-notes/product/2026/august#ai_prep_search-supports-plain-text-and-markdown-input-beta

Related releases