databricks Platform · Updated Jun 11, 2026

ai_extract

Now generally available for structured data extraction.

What it does

The `ai_extract` function enables the extraction of structured data from unstructured text, documents, and images based on a user-defined schema. It supports complex data types including nested objects and arrays, and provides features like type validation, citation tracking, and confidence scoring. This function is integrated with a visual UI for building and refining extraction agents, allowing users to optimize their data extraction processes.

Why it matters

This update enhances data accessibility and usability for organizations by transforming unstructured data into actionable insights, improving reporting and analysis capabilities. It benefits data analysts and business leaders by streamlining workflows, reducing manual data processing time, and enabling more informed decision-making based on structured data extraction.

Key capabilities

Schema-Based Extraction

Extracts data according to a user-defined schema, allowing for tailored data insights.

Visual UI for Agents

Provides a no-code interface to create and manage extraction agents easily.

Support for Nested Structures

Handles complex data types including nested objects and arrays for comprehensive data extraction.

Confidence Scoring

Offers confidence scores for extracted data, enhancing the reliability of insights.

Prerequisites

Workspace with serverless compute enabled.

Unity Catalog must be enabled.

Access to a serverless usage policy with a nonzero budget.

Unstructured data must be in a Unity Catalog volume or table.

Recommended use cases

Extracting legal terms from contracts.Pulling payment details from invoices.Extracting key information from medical records.

Limitations · remember

!Maximum context length for agents is 128k tokens.

!Union schema types are not supported.

Architecture

Data Sources → Unity Catalog → ai_extract Function → Information Extraction UI → Extraction Agents

Get started

Create Agent

Navigate to Agents and select Create Agent > Information Extraction.

Select Data

Choose files or tables from which to extract information.

Define Schema

Configure the extraction schema using natural language or manual entry.

Run Extraction

Save and run the extraction to see results based on the defined schema.

Use Agent

Deploy the agent for SQL queries or create a Spark Declarative Pipeline.

Key takeaway

The `ai_extract` function is now generally available, enabling efficient extraction of structured data from unstructured sources.

2026-06-11-ai-extract-is-now-generally-available Generated by the Databricks Updates Agent · browse all updates