Latest releases from the Databricks ecosystem.

This release adds new resources for Databricks AI Search endpoints and indexes. It also fixes several bugs, including an infinite plan/replace cycle for instance pools and issues with entitlements and MWS resources.

SSH connection error messages are improved with server logs, and the GPU accelerator startup timeout is increased. Bundle deployments now correctly handle Unicode in variable references and prevent drift from backend schema normalizations.

Databricks SDK for Java now supports specifying a serverless compute ID when cloning, creating, or editing Delta Live Tables pipelines. This enables practitioners to manage DLT pipelines with serverless compute configurations programmatically.

The Databricks SDK for Go now includes a ServerlessComputeId field in pipeline-related requests and specifications. This allows users to specify serverless compute for DLT pipelines when cloning, creating, or editing them.

MLflow 3.14.0 introduces new GenAI features like one-command agent setup, durable Claude Code tracing, review queues for traces, and a Pytest integration for regression testing. It also includes a new LLM Playground for prompt iteration and changes the default serialization format for several MLflow models to `skops` or `pt2`.

This release adds new methods for managing Postgres data APIs and introduces fields for Azure compute attributes, synced table specifications, and vector search indexes. A breaking change makes the `resourceId` field optional for bundle deployment operations.

This release adds new methods for managing Postgres data APIs at the workspace level. Compute clusters can now specify an Azure Capacity Reservation Group in their attributes.

This release fixes an issue where tag policy and assignment operations failed with 404 errors when tag keys contained forward slashes. It also introduces a breaking change by making the `ResourceId` field optional for bundle deployment operations.

The `resource_id` field in `bundledeployments.Operation` is no longer required, which is a breaking change. Several bug fixes improve performance and stability, including caching OIDC tokens, making `WorkspaceClient.dbutils` lazy, and better handling of Spark Connect runtimes.

The `database.SyncedTableSpec` and `postgres.SyncedTableSyncedTableSpec` structs now include a `TypeOverrides` field. This allows for specifying custom type mappings when working with synced tables.

The `direct` deployment engine is now Generally Available and the default for new deployments, with existing deployments retaining their current engine. New CLI commands include `databricks quickstart` for an introduction and `databricks version --check` to report available updates.

1.12.1

This release exposes Databricks Jobs IDs in dbt's adapter response, supports SPOG hosts, and fixes issues with streaming tables, materialized views, and Iceberg incremental models. A breaking change requires `contract.enforced: true` for column-level constraints when `use_materialization_v2: true`.

This release adds an `AcceleratedSync` field to `database.SyncedTableSpec` and `postgres.SyncedTableSyncedTableSpec`. This new field enables configuration of accelerated sync for synced tables within Databricks.

This release introduces new services for AI Search and Bundle Deployments, along with numerous new fields across existing services for managing applications, catalogs, ML data sources, and vector search. A breaking change removes the `bundle` package and its associated service.

This release adds new services for AI Search and Bundle Deployments, along with numerous new fields across existing services like Catalog, ML, and Vector Search. The `bundle` package and its workspace service have been removed.

This release introduces new services for AI Search and Bundle Deployments, along with several new fields across existing services like Catalog, ML, and Pipelines. It also includes breaking changes by removing the old Bundle package and its associated workspace-level service.

* Canonicalize Bearer tokenType in Authorization headers

The SDK now correctly handles token-based pagination, preventing silent data loss when an empty page with a next token is encountered. New explicit factory methods for token and offset pagination have been added to the `Paginator` class, deprecating its constructor.

The SDK now better detects AI agents by honoring the Vercel AI_AGENT environment variable and passing through unrecognized agent names in the User-Agent header. This allows more specific agent versions to be reported instead of a generic "agent/unknown".

This release is a backport from the 0.32.x line, which will receive voluntary support for a period. Consult the full changelog for details on specific user-facing changes, fixes, or breaking changes.

* direct: Fix updating the apps after the Go SDK upgrade ([#5444](https://github.com/databricks/cli/pull/5444))

Release: v2.11.0 (#1902)

This release introduces a Unity Catalog explorer and a workspace filesystem explorer for easier navigation. It also adds support for SPOG host URLs.

The experimental open command now supports opening a wider range of Databricks resource types directly in the workspace. Bundles gain a new --select flag for planning and deploying subsets of resources, improved retry logic for transient HTTP errors, and support for Terraform references.

The Databricks Java SDK now correctly handles OAuth token exchanges for browser-based flows where the IdP JWT lacks a client ID, preventing NullPointerExceptions. It achieves this by omitting the client_id parameter from the token exchange request when it's null or empty, enabling account-wide token federation.

Creating an external location with `enable_file_events = false` now correctly sends this setting to Databricks, preventing the server from defaulting it to `true`. This fixes an issue where the `enable_file_events` setting was previously ignored and always defaulted to `true`.

Databricks SDK for Go now includes `DeploymentMode` fields for bundle deployments and versions. It also adds `CollaborationPlatformConnectivity` and `EffectiveCollaborationPlatformConnectivity` fields to settings.

This release adds new `deploymentMode` fields to bundle deployment and version objects. It also introduces `collaborationPlatformConnectivity` and `effectiveCollaborationPlatformConnectivity` fields to the settings API.

This release adds new fields for deployment mode in Databricks Asset Bundles and for collaboration platform connectivity in account settings. These changes provide more visibility and control over bundle deployments and account-level collaboration settings.

MLflow 3.13.0 introduces a new Role-Based Access Control system with an Admin UI for self-hosted instances, alongside trace retention and auto-archival to object storage. It also adds one-click observability for coding agents via the AI Gateway, new engines for MLflow Assistant, and an official Helm chart for Kubernetes deployments.

The SDK now allows setting a discovery account target for login, directing users to the account selector. It also introduces new fields for job and pipeline deployments, token management, and bundle operations.