Skip to main content
Last updated on

Use cases

Region availability

Lakebase Autoscaling is the latest version of Lakebase, with autoscaling compute, scale-to-zero, branching, and instant restore. For supported regions, see Region availability. If you are a Lakebase Provisioned user, see Lakebase Provisioned.

Lakebase Autoscaling supports four primary patterns: serve lakehouse data in Postgres, replicate Postgres back to the lakehouse, run an application backend, and power AI agents and ML. Each pattern uses Postgres alongside Unity Catalog to give your application a low-latency database that stays in sync with the lakehouse.

Serve lakehouse data​

Lakehouse rows synced into Lakebase Postgres for low-latency app reads

Synced tables bring Unity Catalog data into your Lakebase database for low-latency transactional reads. Pick a source table, choose a sync mode, and the pipeline is fully managed. No sync scripts, no external orchestration, no jobs to monitor. Continuous mode keeps data within seconds of the source. Triggered mode balances freshness and cost with scheduled incremental updates. Your application always serves the latest analytics alongside its own operational data.

First steps

Learning path

  1. Create a synced table
  2. Schedule ongoing syncs
  1. Sync modes — Snapshot, Triggered, or Continuous
  2. Data type mapping — UC to Postgres type compatibility
  3. Capacity planning — connection limits, throughput rates
  4. Register in Unity Catalog — cross-source queries across operational and analytical data

Replicate to the lakehouse​

note

Lakehouse Sync is in Beta.

Same row replicated from Postgres to Delta with full change history

Lakehouse Sync continuously replicates your Postgres tables to Unity Catalog as Delta history tables using change data capture. Every insert, update, and delete is captured from the write-ahead log and appended as an immutable row. No external CDC tool, no Spark jobs, no pipeline to maintain. Replication reads from the write-ahead log, not from your tables, so production queries are unaffected. Analytics teams query the full change history from Databricks SQL without touching the operational system.

First steps

Learning path

  1. Replicate to the lakehouse
  2. Create a current-state mirror
  1. How Lakehouse Sync works — CDC mechanism, destination schema, system columns
  2. Data type mapping — Postgres to Delta type support
  3. Managing schema changes — what happens when schemas evolve
  4. Limitations — failure scenarios and troubleshooting

Application backend​

Standard Postgres clients connect to Lakebase Postgres

Your application connects to Lakebase the same way it connects to any Postgres database. Use the drivers and frameworks you already know. When your app gets a traffic spike, autoscaling adds compute without dropping connections. When traffic stops, scale-to-zero suspends the database and reactivates in hundreds of milliseconds on the next query. You don't provision for peak and you don't pay for idle. For development, branching gives every developer an isolated copy of the production database with no data seeding, no storage duplication, and no waiting.

First steps

Learning path

  1. Build a Databricks App — deploy a template with managed credentials
  2. Connect your own application — custom Databricks App, SDK, or API
  1. How autoscaling works — compute behavior under load
  2. Scale to zero — suspension, reactivation, idle timeout
  3. How branches work — copy-on-write isolation for dev/test
  4. Authentication — token rotation for production apps

AI agents and ML​

Agents store memory and models read features through Lakebase Postgres, synced with the lakehouse

Lakebase serves as the backend for AI agent memory and real-time feature serving. Agents built with LangGraph or the OpenAI Agents SDK store conversation state and long-term memory in Postgres. Models served with Mosaic AI access feature data through Online Feature Stores that are powered by Lakebase Autoscaling. Both benefit from automatic scaling, scale-to-zero, and Unity Catalog governance.

First steps

Learning path

  1. Deploy an agent on Databricks Apps — clone a template and deploy
  2. Agent state and memory — add short-term and long-term memory with Lakebase
  1. Stateful agents on Apps — LangGraph checkpointers, memory templates
  2. Stateful agents on Model Serving — time travel, checkpoint management
  3. Feature Store and Model Serving — low-latency feature data powered by Lakebase
  4. Online Feature Stores — create, publish, and manage feature stores
On this page