Databricks SDK

Databricks CLI v1.23.0 makes databricks ssh connect provide claude and codex on dedicated clusters and in serverless IDE sessions and non-interactive commands, and adds GPU_1xRTX6000 workload support plus workspace.sync_mode (overwrite, in-place, immutable), which replaces the deprecated presets.source_linked_deployment and experimental.immutable_folder settings. The release also reverts bundle file uploads to the legacy /workspace-files/import-file endpoint to fix notebooks being uploaded as plain files, adds lifecycle.depends_on and bundle deploy --select for deletions, and fixes bundle validate errors when a schedule, trigger, or continuous block is referenced as a whole.

Databricks SDK for Java v0.172.0 fixes CLI authentication so that small wrapper scripts, such as launcher scripts or version manager shims, are no longer mistaken for the legacy Python CLI, which now only applies to the pip entry point. The release also adds budget spend status methods to the account budgets service and an aiFunctionTask field across the jobs task types, with matching aiFunctionTaskOutput on run output.

Adds get_budget_spend_status() and list_budget_spend_statuses() to the account-level budgets service. Fixes files.download_to with use_parallel=True to download byte ranges from presigned URLs or the storage proxy instead of always using the Files API, and to request the correct path for names containing #, ? or %.

Release v0.191.0 adds SDK methods for AgentKit memory and session store permissions, budget spend status queries, feature engineering materialization sets, and starting or stopping vector search endpoints. It also adds external spend inclusion for budgets, unhealthy-table notifications for data quality, relative error options for approximate count-distinct and percentile functions, and table trigger conditions and timing controls.

Databricks SDK for Java v0.171.0 adds AgentKit permission methods for memory and session stores, Feature Engineering materialization set methods, and new fields across data quality, jobs, ML, pipelines, and sandbox APIs. The release also adds refresh-by-path selection fields to pipeline updates and a fileOutput option on jobs SQL outputs.

v0.157.0 adds Agent Kit methods for granting, revoking, listing, and getting memory and session store permissions, plus create, get, update, and delete methods for Feature Engineering materialization sets, along with new fields for data quality notifications, SQL task file output, pipeline refresh-by-path selection, and sandbox metadata. It also requires certifi 2026.10.19 or newer to pick up an updated CA trust bundle.

Release: v2.22.0 (#2272)

Release v2.22.0 stops bundle sub-project discovery from blocking the extension host and has VS Code handle proxies and certificates for Databricks SDK calls. It also fixes uv workspace member environment selection and constraint conflicts, and exposes the bundle Configuration view and Bundle Resource Explorer in Remote SSH mode.

Release v1.22.0 adds support for the features resource type, lets bundle deployment bind and unbind accept type.key references, and removes the 50-workspace limit in auth login. It also adds DATABRICKS_BUNDLE_STRICT_PROFILE_MATCH for exact profile matching, and release binaries now require macOS 13 Ventura or later.

This release adds start and stop methods for workspace-level vector search endpoints, along with new starting, stopping, and stopped status values. It also adds OneDrive as a pipeline ingestion source, an include_external_spend option for budgets, regex filtering for identity visibility, trigger conditions and timing controls for table triggers, and a relative_error parameter for approximate count-distinct and percentile functions.

Adds SdpSource to ml.DataSource and CustomerVersionLabel, MaxDbu, and ReydenClusterSizeMultiplier to SQL warehouse requests and responses. Breaking: removes ReplicationConfig from sharing.CreateShare and sharing.ShareInfo.

The Go SDK introduces methods for ML experiment issue tracking and sandbox file operations, along with support for Realtime SQL warehouses and job duration targets. Breaking changes remove the EffectiveServerlessComputeId field from job tasks and replace ReadMask with View across Agentkit memory requests.

The SDK adds file and directory management methods to the workspace sandbox service alongside duration target support across Jobs APIs. Unity Catalog connections now support a secrets field, policies include governed tags, and disaster recovery failover groups gain replication tracking fields.

The SDK adds issue tracking methods for ML experiments, a realtime warehouse type for SQL endpoints, and replication configuration support for Delta Sharing. As a breaking change, the effective_serverless_compute_id field was removed from job run tasks.

The SDK adds support for Slack and Microsoft Teams notification destinations, as well as ModelService securable types in Databricks Apps. Model provider configurations in Unity Catalog now include pricing fields and support for Bedrock Mantle.

The SDK adds account-level management for identity visibility filters and introduces statement timeout configuration for SQL warehouses. It also adds parent path support for pipelines, custom template formats for SQL alerts, and streaming shuffle partition settings for machine learning features.

This release adds the Agent Kit workspace service, account IAM identity visibility filters, and statement timeout configuration for SQL warehouses. It also introduces a breaking change by removing the Mason workspace service and package.

The Terraform deployment engine for Databricks Asset Bundles has been removed, requiring users to migrate to the direct engine or remain on CLI v1.19.x. Additional updates include a fix for large integer corruption in API commands, group-level run-as support in bundles, and expanded authentication options for Databricks Apps.

The SDK now includes the AgentKit package and workspace service, as well as a HealthCheck field for app deployments. In a breaking change, the Mason package and its workspace service have been removed.

MLflow 3.17.0

MLflow 3.17.0 introduces fine-grained access control for runs and traces, links evaluation metrics directly to managed datasets, and adds support for Databricks Unity Catalog model service resources. Self-hosted SQL tracking servers require a coordinated schema migration via mlflow db upgrade with all writers stopped, as rolling deployments are not supported.

A new account-level service is now available to manage private network gateways. Additionally, the SDK adds a project environment field for compute environments and a dataframe schema field for machine learning request sources.

The Go SDK now includes the PrivateNetworkGateways account-level service for network management. Additionally, a new DataframeSchema field is available for machine learning request sources.

The SDK introduces environment variable configurations and maintenance window notifications across Jobs and Tasks, along with new skill management APIs in AI Gateway. It also adds the Mason workspace service and makes CommandPath in job deployments and ApiSecretRef in ML schema registries optional.

The Databricks Python SDK introduces a new list_commands method to the workspace sandbox service. This update allows practitioners to programmatically list and inspect commands within Databricks sandbox environments.

The AI functions service now supports an ai_decide method, and multiple Jobs and Tasks models now accept environment variable configurations. Additionally, command_path in DeploymentSpec and api_secret_ref in SchemaRegistryConfig are no longer required fields.

The Databricks SDK for Go now supports Config.Headers to configure custom HTTP headers on all client requests in addition to normal authentication. Practitioners can use the new StaticHeaders helper to easily apply a fixed set of headers.

The Databricks Python SDK introduces skill management operations to the AI Gateway workspace client. Users can now create, retrieve, list, update, finalize, and delete skills programmatically using new methods on the AI Gateway service.

AI Runtime commands have moved from experimental to databricks air, and SSH commands now support keeping detached background processes running after tunnel disconnect. Databricks Asset Bundles direct engine resolved multiple issues around unnecessary resource recreations, Unity Catalog grant convergence, and Git-sourced Python tasks.

MLflow 2.11.5

Users can now opt in to route Unity Catalog model registry artifact uploads and downloads through the Databricks SDK Files API. This update provides an alternative transfer mechanism for managing model artifacts within Databricks environments.

This release adds support for the new Mason workspace-level service via the w.mason client. It also introduces service_credential and secret_reference fields to Unity Catalog model provider configurations.

This release adds AI Gateway credential management APIs for MCP services and expands data pipeline connectors to include TikTok Ads and Smartsheet as ingestion sources. It also introduces Avro and Protobuf transformer format support for pipelines, timezone configuration for ML job scheduling, and new securable types for agent services and skills.

The SDK adds new credential management methods for MCP services in AI Gateway and expands job notifications to include maintenance-related events. Support for additional data formats (Avro, Protobuf) in pipelines and new connection types (TikTok Ads, Smartsheet) extend integration capabilities.

Direct deployment state version 3 is a breaking change requiring Databricks CLI v1.8.0 or later. New capabilities include Docker credential helpers for Databricks Artifact Registry, AI Gateway service support in bundles, file-change job run triggers, and bundle improvements for resource reference handling, deployment reporting, and state management.

Add GroupName field for pipelines.RunAs.

The SDK adds support for new fields across multiple services: AI Runtime jobs can now specify image paths and priority classes, ML models support budget policies and tags during publishing, and serving configurations gain workspace credential options including new GPU accelerator types. Pipelines add development mode configuration and workspace settings gain ingress network policy controls.

SSH sessions now automatically reattach and replay missed bytes when the tunnel connection drops temporarily, preventing session interruptions and eliminating manual reconnect workarounds. The CLI adds PyDABs secrets support, JSON output for aitools install with error categorization for CI automation, and displays bundle sync progress by default.

Release introduces three new workspace services (AI Functions, Domains, Sandbox), extends Feature Engineering with new backfill and operation management methods, and adds AWS Secrets Manager and Azure Key Vault connection support. Multiple breaking changes remove fields from model serving and catalog configurations that may require code updates.

The SDK adds a purge_feature_entities() method for feature engineering and introduces new deny policy type support with a deny field in PolicyInfo. New budget_policy_id and tags fields are available for IngestionConfig and MaterializedFeature to enable budget policy tracking and resource tagging.

The SDK adds a new domains workspace-level service for managing domains. ML Feature Store capabilities are expanded with new fields for feature filtering, scheduling, and improved source column specifications across Delta tables and Kafka sources.

The SDK adds an execute_command_sync() method for sandbox and new fields for ML features (mode, time_window, full_feature_name). Breaking changes remove routing and rate limiting configuration fields from model serving APIs.

The Pipelines API's source_table and source_schema fields are now optional to support streaming and message-bus connectors that don't use source tables or schemas. New ai_functions and sandbox workspace services are available, along with expanded APIs for bundle deployments, feature engineering, PostgreSQL, IAM v2, and model deployment resources.

This release adds support for Postgres snapshot operations, RabbitMQ pipeline ingestion, external IAM entity lookups, and H100 GPU compute. It also introduces breaking changes by removing several ML service fields and making source schemas and tables optional in pipeline specifications.

Bundles automatically migrate from Terraform state to the direct engine on clean deploys (opt-out via engine: terraform), and SSH sessions stay connected during idle periods with automatic keepalives every 30 seconds. Direct engine bundle operations now correctly handle removed configuration fields, support cluster policies resources, and include a DATABRICKS_BUNDLE_RESOURCE_MAX_WAIT timeout setting.

Databricks Asset Bundles now report detailed resource action summaries and file sync counts during deploy and destroy operations, alongside fixes for workspace Git folder metadata and state migration. The CLI now enables FIPS 140-3 compliance by default for TLS connections, adds Goose support to aitools, and automatically resolves conflicting databricks-connect pins in local environment setups.

Add EnabledTelemetryFeatures field for serving.TelemetryConfig.

Fixes OAuth token cache conflicts in multi-process deployments and Spark runtime version handling, with breaking changes requiring code updates for pagination in IAM list operations. Introduces git-based deployment for apps and custom UDF support in ML functions.

Go version requirement raised to 1.25, and IAM list methods now include pagination with breaking signature changes, while CreateOperation is removed from BundleDeployments. Fixes concurrent OAuth token cache handling and Spark runtime selection for major-only versions.

The SDK makes action_type and status fields optional instead of required in bundle deployment operations. The create_operation() method is removed from the bundle_deployments workspace service.

The SDK adds LinkedIn Ads and Marketo connector options plus PostgreSQL halfvec and varchar type support for synced tables. AccountIamV2API.list_workspace_assignments and WorkspaceIamV2API.list_workspace_assignments_proxy now require pagination handling, introducing breaking changes that may affect existing code.

The SDK adds pagination to four IAM methods (list_direct_group_members and list_workspace_assignment_details in both account and workspace APIs), which is a breaking change requiring code updates. New fields are added for Genie conversation analysis, group membership tracking, serverless compute identification, and Postgres pipelines.

The SDK adds new methods for managing groups, users, service principals, and workspace assignments at both account and workspace levels, providing expanded IAM capabilities. The NewCluster field in JobCluster is now optional, a breaking change that affects existing cluster configuration code.

The SDK adds comprehensive IAM v2 methods for managing users, groups, service principals, and workspace assignments at both account and workspace levels. The new_cluster field in JobCluster is no longer required, which is a breaking change.

Bump github.com/databricks/databricks-sdk-go from v0.166.0 to v0.170.0 (#6251).

The CLI adds databricks environments setup-local to provision matched Python environments for Databricks compute targets and extends aitools install to support Gemini CLI and Pi. Bundles fix the ignored bundle.deployment.lock.force setting, add pipeline cascade_on_destroy control, improve experimental job_runs with idempotency tokens and completion waiting, and add UC secrets resource support.

The SDK adds job triggers for scheduling, API source connectors for pipelines, secret value retrieval in requests, new GpuXlarge8 workload types, Netsuite connection support, and token forwarding for apps. Breaking changes remove browse-related fields from model and MCP service APIs and make several user and service principal fields required.

The Terraform deployment engine is now deprecated; migrate to the direct deployment engine for continued support. AI runtime tasks now automatically package local directories into tarballs during deployment, and the direct engine fixes several convergence issues with empty grants, webhook notification ordering, pipeline configuration, and vector search index creation.

The Databricks SDK adds principal-level overrides and resource-type tracking to billing configurations, plus a new block_usage billing action type. New features include NetSuite connection support in the catalog, grant-level policy control, and input/output column mapping for pipeline transformers.

The Go SDK adds fields for granular billing control: PrincipalOverrides, ScopeType, and ResourceType for alerts and budgets, plus a new BlockUsage action type for alert configuration. The catalog API now supports grant management for policies via a new Grant field and PolicyTypeGrant enum type.

Breaking changes include removal of direct AWS and Azure credential fields from model provider configurations and State field type changes in bundle deployment operations. New features add KinesisStreamConfig support for ML streaming sources, a Mode field for pipeline updates, and Pi coding agent attribution in the User-Agent header.

The SDK adds new fields to job triggers (PauseStatus), workspace objects (DirectoryInfo), and IAM entities (GroupId, ServicePrincipalId, UserId) for enhanced API access. A breaking change removes the InternalId field from IAM Group, ServicePrincipal, and User types, requiring code updates for any existing consumers.