Databricks Genie Agents Expand with Unstructured Data Analysis and Native MCP Connectors

Databricks Genie Agents now support unstructured data analysis in Public Preview, enabling queries across thousands of documents (PDFs, slides, images) in Unity Catalog volumes simultaneously with governed tables.
Native MCP connectors are Generally Available, supporting tools such as Jira, Confluence, Slack, Google Drive, and Microsoft 365 for context reading and actions like ticket creation and message posting.
OpenSharing and Genie Clean Rooms are in Beta, allowing the sharing of fully functional agents across Databricks accounts and enabling privacy-focused collaboration on sensitive data without exposing underlying datasets.
Accuracy is maintained through a business context layer that learns definitions and metrics, with benchmarks used as unit tests for regression detection and production monitoring for quality improvement.

Unstructured Data Analysis

Uses a custom ai_parse_document function to recognize tables, figures, and headers while preserving context.
Provides page-level citations and enforces Unity Catalog permissions so users only access files they are authorized to view.
Reduces the need for custom RAG pipelines, separate AI Search endpoints, and embedding pipelines.

Lakeflow Connect collects data from Salesforce, Workday, SAP, and Marketo into Unity Catalog for analysis.
All MCP connectors are governed through Unity Gateway, allowing agent authors to set per-tool access permissions.

OpenSharing allows data providers to package agents on Databricks Marketplace for data intelligence sales.
Genie Clean Rooms include controls such as conversation limits and output size restrictions to protect intellectual property.