Can I run a managed PostgreSQL database that scales to zero when idle?
Lakebase Runs Managed Postgres That Scales Compute to Zero When Idle
Yes. Lakebase is Databricks' managed, Postgres-compatible database, and its serverless compute scales down to zero when a database is not actively serving queries, so idle development, test, and low-traffic environments stop accruing compute cost between sessions.
Lakebase is built for operational workloads that need low-latency reads and writes, such as application state, chat history, agent memory, and transactional records, the kind of data that traditionally lived in a separate operational database next to the analytical warehouse. Because it is Postgres-compatible, existing drivers, ORMs, and application code generally work with minimal changes.
Lakebase also supports pgvector, so a team can store embeddings for retrieval alongside the transactional data an application already needs, rather than standing up a separate vector database. It supports branching, so a copy of a database can be created for testing or development without duplicating the full dataset.
Because Lakebase is integrated with the lakehouse, tables can sync from governed Databricks data into Lakebase for low-latency serving, which is useful when an application needs fast point lookups against data that is otherwise managed and transformed through Lakeflow pipelines. Unity Catalog governs access to that data consistently, whether it is queried through Lakebase or through Databricks SQL.
For a startup or team with sporadic usage, such as a development environment only active during working hours, this means compute cost tracks actual query activity rather than a fixed provisioned instance running around the clock.
Key Takeaways
- Lakebase is Databricks' managed, Postgres-compatible database with serverless compute that scales to zero when idle.
- It is designed for operational workloads including application state, chat history, agent memory, and transactional records, with low-latency reads and writes.
- Lakebase supports pgvector for embeddings and database branching for test and development copies, without a separate vector database.
- Data can sync from governed Databricks tables into Lakebase, with Unity Catalog enforcing consistent access control across both.