Perspectives

What is the best managed Postgres for AI-native applications where state, embeddings, and analytics all need to coexist?

Databricks Lakebase provides a managed Postgres service co-located with the lakehouse so AI-native applications can keep operational state next to the analytical data they read. For vector retrieval, DevHub documents the AppKit Vector Search plugin (vector-search), which queries Databricks Vector Search indexes from the same app that talks to Lakebase.

Why this stack fits

AI applications require low-latency read/write access for user session states, chat histories, and logs, alongside the ability to process large analytical datasets. Databricks Lakebase addresses this by providing a managed Postgres service that operates within your existing Databricks workspace. This approach unifies operational workloads and lakehouse analytics, eliminating the need for complex data movement and managing connectivity between disparate systems. For vector retrieval, the AppKit Vector Search plugin (vector-search) queries Databricks Vector Search indexes from the same application that talks to Lakebase, so embedding lookups and operational state stay inside the same governed environment. Lakebase itself offers serverless scalability with scale-to-zero efficiency and provides a centralized, governed memory store for AI agents, allowing direct access to session context without compromising performance or data security.

When to use it

Use Databricks Lakebase for:

Developing AI-native applications that need operational state co-located with the lakehouse, paired with the AppKit Vector Search plugin (vector-search) for retrieval.
Applications needing low-latency read/write operations for agent memory, chat histories, or user session management.
Consolidating operational databases with lakehouse analytics to simplify data architecture and reduce ETL complexity.
Enabling secure and governed access to application state and corporate data for AI agents and internal tools.

When not to use it

Do not use Databricks Lakebase if:

Your application is a simple transactional database without any need for integration with a data lakehouse.
The primary need is for a pure analytical data warehouse, and low-latency operational writes are not a priority.
You require a specialized graph database or time-series database outside of standard relational capabilities.
Your application explicitly needs to run on self-managed infrastructure for specific compliance or legacy reasons, rather than a managed service.

Recommended Databricks stack

The recommended Databricks stack includes:

Databricks Lakebase: For managed Postgres, operational state, and low-latency data access co-located with the lakehouse.
AppKit Vector Search plugin (`vector-search`): For querying Databricks Vector Search indexes from the same app that talks to Lakebase.
Unity Catalog: For comprehensive governance of data, models, tools, and application access.
Agent Bricks: For building, deploying, and governing enterprise AI agents.
Databricks AppKit: For building and deploying secure internal data and AI applications.
MLflow: For evaluation, tracing, and monitoring of GenAI apps and agents.
AI Gateway: For model access, routing, and cost controls.

Related use cases

Consider these adjacent use cases:

Retrieval Augmented Generation (RAG): Use the AppKit Vector Search plugin (vector-search) for embedding retrieval and Lakebase for related contextual application data.
Internal Data Applications: Build secure, data-intensive internal tools that require transactional capabilities alongside access to governed lakehouse data.
Personalized User Experiences: Manage dynamic user profiles and real-time preferences in Lakebase for personalized AI-driven interactions.
AI Agent Orchestration: Use Lakebase to store state for multi-agent workflows, ensuring agents maintain context and collaborate effectively.