What is Lakebase Autoscaling?

Lakebase Postgres Autoscaling is a fully managed Postgres database integrated into the Databricks Data Intelligence Platform. It is built for any application that requires online transaction processing (OLTP) and low-latency data serving. Lakebase brings these capabilities to your lakehouse, enabling you to build real-time transactional applications alongside your analytics workloads.

Lakebase Postgres Autoscaling combines the reliability and familiarity of Postgres with modern database capabilities including autoscaling, scale-to-zero, branching, and instant restore. These features enable flexible development workflows, cost-efficient operations, and rapid iteration.

The diagram shows how Lakebase integrates with the rest of the platform: real-time feature serving for ML models and Feature Store, agent state for AI agents, and transactional data for Databricks Apps or any application you connect to it.

You can move data in either direction between your lakehouse and Lakebase. Synced tables move data from the lakehouse into Lakebase so your applications can query it at low latency.

Example use cases and workload types

The following are just a few examples of the many ways you can use an OLTP Postgres database like Lakebase across industries: personalized recommendations and offer targeting in e-commerce and retail, clinical trial data and recommendation systems in healthcare, automated trading and streaming analytics in financial services, and machine telemetry and maintenance workflows in manufacturing.

Common workload types for OLTP databases may include the following:

• Data serving: Serve insights from golden tables to applications at low latency and high QPS.
• Store application state: Manage workflow and agent state in a transactional data store.
• Feature serving: Serve featurized data at low latency to ML models.

Databricks integration

The diagram above highlights three key integration use cases:

• Real-time feature serving: Use Lakebase projects as an online store for ML models and Feature Store, so you can serve featurized data at low latency. See Online Feature Store (Lakebase) and Feature Serving.
• Agent state for AI agents: Store and manage state for AI agents in a transactional database, so conversations and workflow context persist across requests.
• Transactional data for applications: Persist data for Databricks Apps or any application you connect to Lakebase. For Databricks Apps, add a Lakebase project as an app resource. See Add a Lakebase resource to a Databricks app.

Lakebase Provisioned

Lakebase Provisioned is the original Lakebase offering that uses provisioned compute you scale manually. Existing Provisioned instances continue to be supported. New Lakebase development is focused on Autoscaling. If you have Provisioned instances or are evaluating both options, see What is Lakebase Provisioned? and Autoscaling by default.

What is a project?

Lakebase Autoscaling resources are organized into a project structure. A project is the top-level container for your database resources. When you create a Lakebase Autoscaling database, you create a project. The project holds your branches (database environments), computes, roles, and databases. Think of a project as the unit of organization for one application or workload. You can have multiple projects in a workspace, each with its own branches and data.

How projects are organized

Understanding the hierarchy of objects within a project helps you organize and manage your resources:

Databricks Workspace
  └── Project(s)
      └── Branch(es)
          ├── Compute (primary R/W)
          ├── Read replica(s) (optional)
          ├── Role(s)
          └── Database(s)
                 └── Schema(s)

Each level in the hierarchy serves a specific purpose:

Understanding branches

One of Lakebase Postgres's most powerful features is branching. Like Git branches for your code, branches let you create isolated database environments for development and testing—without affecting production.

Why this matters: Traditional database workflows require separate dev and staging servers, manual data refreshes, and careful coordination. With branches, you can:

• Instantly create a development environment with production data
• Test schema changes safely before applying them to production
• Recover from mistakes by creating branches from any point in time
• Pay only for the data you change, not full duplicate databases

Core concepts

Lakebase is built on several key innovations that differentiate it from traditional database systems:

• Separated compute and storage: Scale compute resources independently from storage for cost efficiency and flexibility.
• Autoscaling: Compute automatically adjusts based on workload demand, with support for scale-to-zero during idle periods.
• Copy-on-write storage: Enables instant branching where you only pay for data changes, not full duplicates.
• Instant point-in-time operations: Create branches or restore to any moment within your configured restore window (0-30 days)

These concepts work together to enable flexible development workflows, cost-efficient operations, and rapid recovery from mistakes.

For a detailed explanation of each core concept, see Core concepts.