Skip to main content

Platform overview

Databricks Apps, Lakebase Postgres, Agent Bricks, and the Data Lakehouse are the four core workspace services that make up a full-stack Databricks application. The diagram below shows how they fit together.

Architecture diagram of the Databricks workspace: Databricks Apps contains AppKit, whose Lakebase, Model Serving, and Analytics plugins connect to Lakebase, Agent Bricks, and the Data Lakehouse as sibling services inside the workspace; a fourth column lists additional AppKit plugins (Server, Genie, Files, Jobs, Vector Search, and Custom Plugins)
  • Databricks Apps: managed hosting for apps you build with AppKit, the TypeScript SDK for production-ready Databricks applications.
  • Lakebase Postgres: managed Postgres database for OLTP storage co-located with your workspace data. Use it for sessions, app state, conversation history, or any data your app reads and writes at low latency.
  • Agent Bricks: Databricks' enterprise agent platform, unifying model access, execution, governance, and context. Use it for AI features in your app: chat with your company's docs (Knowledge Assistants), ask-your-data in plain English (Genie), foundation-model calls, or custom Python agents.
  • Data Lakehouse: governed analytical data in Unity Catalog. Use it to read company data, trigger Lakeflow Jobs, and display data freshness in your UI.

How a request flows

  1. A user opens the app at its workspace URL. Databricks Apps authenticates them via workspace SSO.
  2. Each AppKit plugin then handles requests for its service: the Lakebase Plugin queries Lakebase, the Model Serving Plugin calls Agent Bricks, the Analytics Plugin reads the Data Lakehouse.
  3. Each plugin call runs as the app's service principal (by default) or the user's forwarded token (when per-user permissions matter). Workspace permissions and governance apply automatically.

Where to next