5.33 Release Notes - Databricks

New Features

Cortex and Fabric Data Agents for AI Query

EARLY ACCESS

AI Query now extends beyond Databricks Genie to Cortex and Fabric Data Agents, using your Snowflake and Power BI connectors respectively.

AI Query lets a provider provision access to an agent they have already built, directly through the Harbr UI. Questions and answers come back as text and table responses.

• Providers register an agent by choosing a connector, selecting from the agents that connector can reach, and publishing.

• Consumers with the 'access' permission on a product can ask questions in natural language. They do this in a chat window that opens from the product page and from My Collection.

• No credentials are shared and no data is moved onto the platform. Every question goes through Harbr using the connector's identity, and each one is recorded so usage and cost are visible.

For user guidance, see AI Query Assets.

AI Query asset MCP tool

EARLY ACCESS

Your AI assistant can now work with AI Query assets, using the model of your choice, through the new AI Query tool in the Harbr MCP. This allows you to gain insight and value from the data the AI Query asset has access to all from within your existing AI chat interface. It sits alongside the existing subscribe, export, and data-sharing tools.

For user guidance, see the update to Using the Harbr MCP.

File assets in Spaces

You can now work with file assets in Spaces in three ways:

• Bring existing file assets into a Space

• Create files within the Space

• Create new file assets from a Space

File assets in Spaces only work in Spaces created after the feature is enabled, so existing Spaces need to be recreated to use it. Harbr handles the setup directly, so there is nothing to configure on your side.

For user guidance see Space and File Assets in Spaces.

Improvements

Serverless compute everywhere

Asset creation, updates, and export can now run on Databricks serverless compute. This removes the dedicated cluster start-up time of five to six minutes or more from every job. Asset creation, update, and export are all much quicker as a result, especially for smaller jobs.

• Turn it on with a single platform-level toggle. There's no need to configure each connector type separately.

• Once enabled, eligible connectors use serverless automatically. There's no separate opt-in step per connector, and all existing journeys behave exactly as they do today.

• To use serverless, run a capability check on your connector to confirm serverless compute works with your downstream journeys.

• You can opt individual connector types out of serverless if you need their processing to stay on dedicated compute.

• We've also upgraded the Databricks engine components used by serverless and tuned their behaviour, to further optimise performance.

For operator set-up, see:
Configuring Serverless Compute for Asset Creation and Export

Capability checks now gate serverless vs. classic compute automatically

Capability checks now decide whether a connector uses serverless or classic compute. This catches a connector that would fail on serverless before you rely on it.

• In scenarios where both are configured, each connector is tested both ways. If the serverless check fails, the journeys using that connector will automatically fall back to classic compute

• Connector creation succeeds as long as one of the two checks passes, so a serverless-only issue won't stop you creating the connector.

• Once you've fixed the underlying issue, re-run the capability check and, if successful, the journeys using that connector will make use of serverless compute

• An event is raised whenever a serverless check fails, so your operations team can see it happening.

For user guidance, see the relevant section within Create a Connector.

For operator set-up, see Configuring Connector Capability Checks and Capability Checks for Serverless

Repeating export to desktop download

EARLY ACCESS

You can now set an export with a desktop target to repeat, so the latest version of your data is always ready to download. Desktop downloads now have their own dedicated experience within export, optimising the UI experience when download is the target

• Set-up is unchanged: choose Desktop as the export target and select Repeating.

• The export shows when the data was last refreshed and when you last requested a download. It also flags when a newer version has become available since your last download.

• The Updates view shows each version's refresh time and download request time, so you can see at a glance which versions you have and haven't downloaded.

• Please note that you are only ever able to download the most recent version of the data. This can be done from within the Export or directly from the export list using the “…” menu.

For user guidance, see Repeating Export to Desktop Download.

Optimised OneLake Asset creation

Asset creation from OneLake connectors can now read large tables in parallel chunks with stable boundaries, to increase stability and performance.

• Enabled per OneLake connector, so it can be switched on in the individual connector settings using the metadata key value pair - Key: enablePartitionedRead Value: true

SEO improvements to Open Pages

Public Pages are now built to be indexable, enabling Organic acquisition. Making public pages findable and searchable as they are able to work properly with search engines and link sharing - driving Top of Funnel.

• Search-ready product pages and open pages

• New Search and discoverability section in open-page editor

• Faster, fully-crawlable hubs

• Edits go live faster

For full detail, see Open Pages.

Delete users who were invited but never accepted

You can now delete a user who was invited but never accepted or registered, and re-invite them if needed. This builds on user deletion, introduced in 5.32.

See Delete a User

Fixes

• Scheduled asset updates are now more reliable.

• Scheduled updates now run for assets sourced from Azure, Amazon S3 and Google Cloud Storage connectors.

• Updates to assets that have not yet been released now run reliably.

• File-triggered updates now run for table-based assets created from an SFTP connector.

• Asset updates from Microsoft OneLake and Fabric sources now recover automatically if the source connection drops during a read.

• Automatic timeouts for asset updates now apply from an asset's first successful update, so a stalled update is stopped promptly even on new or infrequently refreshed assets.

• Queries issued by Harbr now appear as Harbr in your Databricks Query History, rather than as "Unknown".

• Column descriptions now appear alongside column names and data types when selecting columns on the Export screen.