Lakebase Postgres

Lakebase Postgres is the clearest example of a lakebase. To your application it behaves like standard Postgres. Its storage layer has been rebuilt so that durability lives in cloud object storage instead of on a disk attached to a single machine.

The architecture

The engine splits into four roles, and only the first one runs your queries:

• Compute is stateless and ephemeral. It runs the Postgres you already know: parsing, planning, transactions, MVCC, indexes. It keeps a local cache in RAM and NVMe for speed but owns no durable state, so it can start, stop, scale, or fail without losing data.
• Safekeepers replicate the write-ahead log (WAL) across nodes using a Paxos-style consensus. A commit is defined by quorum acknowledgement, not by a flush to one local disk.
• Pageservers rebuild any page on demand by replaying the WAL up to a requested log position (LSN). This work happens off the hot query path.
• Object storage holds the immutable, versioned history of every page plus database snapshots. It is the source of truth for durability, and it is not on the critical path of a live query.

Because history is retained in object storage and compute holds no durable state, several features that are expensive in a traditional database become cheap metadata operations here.

What it enables

• Scale to zero. Idle compute suspends and reactivates in milliseconds, so a mostly idle database costs almost nothing to keep around.
• Autoscaling. Compute grows and shrinks with load, with no restarts.
• Instant point-in-time recovery. Restore to any moment in the history window without copying data.
• Copy-on-write branching. Create a full, isolated clone of the database instantly. Parent and branch share storage until they diverge, so a fleet of thousands of branches consumes very little extra space.
• Read replicas and connection pooling that read from the same shared storage.

These properties suit AI agents and app-generation platforms, which spin up large fleets of small, mostly idle, frequently branched databases.

One product, two access paths

Lakebase Postgres is a single database product with the same engine and the same core features regardless of how you reach it:

• Via Neon, it is aimed at developers, startups, and agent platforms. It comes with a Free plan and sits inside Neon, a complete set of cloud backend primitives built around Lakebase Postgres (Object Storage, Functions, Managed Better Auth, and an AI Gateway).
• Via Databricks, it is aimed at enterprises, integrated with the Databricks Platform, Unity Catalog governance, and the lakehouse.

Branching, autoscaling, scale to zero, read replicas, and instant restore are available on both access paths. Neon's storage engine is open source.