Session

Lakebase: Building the Operational Foundation for AI Agents

Overview

ExperienceVirtual
TechnologiesLakebase

Lakebase is a new approach to radically simplify how organizations operate. It's a new category of OLTP databases based on Postgres that allows users to seamlessly integrate real-time OLTP workloads with the Lakehouse. Hear from Andre Landgraf, a key voice of the Lakebase community on the technology behind it all, and how to get started building.

Session Speakers

Andre Landgraf

/Staff Developer Advocate
Databricks

Speaker placeholderIMAGE COMING SOON

Jason Pohl

/Head of BuiltOn Field Engineering
Databricks

Maria Zervou

/Chief AI Officer - EMEA
Databricks

Full Summary

Why Lakebase brings serverless Postgres to the Databricks Platform

Lakebase brings fully managed, serverless Postgres to the Databricks Platform, closing long-standing gaps between operational applications, agent workloads, and analytics. The pitch is straightforward: keep the Postgres interface developers already use, then re-architect beneath it so compute and storage scale independently. That unlocks near‑instant database startup, autoscaling, scale to zero when idle, and database branching that fits modern development and agent workflows. Teams exploring how AI is transforming data analytics will find that Lakebase removes a persistent bottleneck: the gap between where operational data lives and where analytical and agent workloads need it.

FAQ


Yes. Lakebase is Postgres compatible, so existing connection strings, drivers, and patterns work. Many popular extensions, including pgvector, are supported. Each new extension must be made available and maintained on the platform, so availability is based on demand and evaluation.

Lakebase can bring up a new Postgres instance in less than a second and wake just as quickly after scaling to zero. Autoscaling responds to workload needs. Scale to zero is configurable and best for dev or idle branches, while production environments can keep compute warm for instant responses.

No. Branches are instant, copy‑on‑write clones of both schema and data, but they do not auto‑merge. Apply any schema changes to the production branch through your normal migration process after reviewing branch outcomes.

Operational data lives in Postgres for app responsiveness, while the platform reduces the need for custom syncs to the lake for analytics. Teams can run heavy analytical queries on lake data without overloading the operational database, shortening the loop from writes to insights.

Two ways stand out. Coding agents can provision and manage an app's database using fast spin‑up and branching. Agents can also use Postgres as memory and state, with pgvector enabling semantic search over context, all while keeping experiments isolated from production.