Databricks Asset Bundles

Overview

Databricks Asset Bundles package a Databricks project as code: jobs, pipelines, dashboards, model serving endpoints, experiments and registered models, defined in YAML and deployed with the Databricks CLI. Databricks renamed them Declarative Automation Bundles in 2026, though the original name is still what teams say.

They are scoped to one platform, deliberately. A bundle deploys objects inside a Databricks workspace, not cloud infrastructure. For workspaces and accounts themselves, Databricks points you at its Terraform provider instead.

The workflow is validate, plan, deploy, run. State is kept in the workspace file system rather than on your machine, so a team deploys against the same tracked resources, with locking to prevent clashing deployments.

Targets give per-environment deployments from one definition. Development mode prefixes and tags resources per user and pauses schedules, so a developer's copy cannot fire on a schedule. Production mode adds guardrails, including checking that you are on the expected Git branch.

One thing to know if you read older material: bundles were originally built on the Databricks Terraform provider, but a direct deployment engine is now the default for new bundles and Databricks says the Terraform engine will be disabled. Descriptions of bundles as "Terraform underneath" are out of date.

Existing workspace objects can be adopted rather than rebuilt: the CLI can generate configuration from them and bind definitions to what already exists.

Key features and capabilities

The same headings are used for every infrastructure as code entry, so two tools can be read side by side.

How infrastructure is described

• Declarative YAML, not HCL or a general-purpose language
• Top-level sections for bundle, variables, workspace, artifacts, resources, targets and permissions
• Reuse through file includes, variables with lookups, and per-target overrides

State and planning

• State lives in the workspace file system, so the team shares one view of deployed resources
• A plan command shows what will change before deploying
• Deployment locking prevents concurrent deployments clashing
• Existing resources adopted by generating configuration, or binding definitions to them

What it can provision

• Databricks workspace objects only, around 38 resource types
• Jobs, pipelines, dashboards, model serving, experiments, catalogs, volumes and secrets
• No clouds, SaaS or Kubernetes; the Terraform provider covers Databricks infrastructure

Policy, testing and secrets

• Permissions set by group, user or service principal, with run-as for the executing identity
• Secrets and secret scopes are first-class resources
• Production mode checks the Git branch and blocks development settings
• No policy-as-code engine comparable to Sentinel or CrossGuard

Team workflow

• Targets give development, staging and production from one definition
• Development mode prefixes and tags resources per user, and pauses schedules
• A summary command prints deep links to deployed resources
• Reuse is through includes and bundle templates rather than a module registry

Integrations

• Documented CI systems are GitHub Actions, Azure DevOps and Jenkins
• Service principals recommended for CI rather than user accounts
• Git metadata captured in the bundle definition

How it runs

• CLI only; the control plane is the Databricks workspace, so nothing to self-host
• Deployment engine selectable, with direct the default for new bundles and Terraform being retired
• Proprietary Databricks Licence, tied to a Databricks subscription

Pricing

Free

No separate charge is published: bundles and the CLI are not line items on the Databricks price list, and use is tied to a Databricks subscription. Databricks itself is billed by consumption, with committed-use discounts quoted privately. Unlike Bicep, Databricks publishes no explicit statement that bundles are free.

Vendor pricing page →

Demos and videos

• Databricks Asset Bundles: a unifying tool for deployment on Databricks Product demo · YouTube

About Databricks

Bundles are a feature of the Databricks platform, delivered through the Databricks CLI, which is published on GitHub under the proprietary Databricks Licence rather than an open-source one: it may be used only in connection with a Databricks subscription. Development is Databricks-led with no foundation governance, and there is no separate commercial edition of bundles. The CLI reached version 1.17 in September 2026.

databricks.com