Embedded Analytics with Databricks AI/BI

It also situates the approach in the broader move to AI-powered self-service analytics that shortens time to value. Teams exploring the full scope of embedded intelligence will find useful context in the complete guide to business intelligence and analytics in the AI era.

The conversation centers on three outcomes embedded analytics consistently deliver: net-new revenue from premium tiers, product differentiation, and faster time to market by avoiding custom charting and governance rebuilds. The emphasis is on concrete, repeatable patterns for permissioning, multi-tenancy, and authentication rather than bespoke builds.

Why embed analytics on the Databricks Platform

Keeping everything Lakehouse-native avoids parallel stacks. Unity Catalog permissions, semantic definitions, and governance flow directly into embedded surfaces, which removes the need for a second semantic layer or contract.

Performance and scale travel with you. Caching, materialized views, and client-side filtering that serve internal users also accelerate embedded experiences. Pricing is usage based, with no per-user or per-embed licensing, which is often decisive compared with traditional BI tools.

External AI/BI Dashboard embedding at scale

For audiences without Databricks accounts, use external embedding backed by a service principal. The application authenticates the end user, retrieves service principal credentials from a secret manager, exchanges them for an OAuth token, and includes a custom claim that identifies the tenant. The AI/BI embedding library consumes the token. Unity Catalog applies row-level security keyed to that claim, so one dashboard definition safely serves many tenants.

Design choices matter. Favor individual data permissions that reflect the viewer's entitlements over shared permissions inherited from an author. For multi-tenancy, mix a common, reusable dashboard secured by row filters with tenant-specific content only where truly needed, which avoids duplicating artifacts.

Implementation is straightforward: data governed in Unity Catalog, Databricks SQL, an AI/BI Dashboard, a service principal with the right privileges, and a lightweight backend and frontend to mint tokens and render the embed. A customer example highlighted production rollout in about a week, with measurable retention lift among feature-activated accounts.

Going beyond dashboards with Genie spaces and APIs

Genie integrations range from no-code to fully programmatic:

• Iframe embedding, best for internal users who can authenticate to Databricks.
• Genie Conversation APIs, which provide low-level control to start conversations, send follow-ups, poll for responses, fetch SQL results, and re-execute cached queries.
• MCP servers, with two flavors: Genie Spaces MCP (single space, stateless, no conversation history) and Genie 1 MCP (stateful, spans all Genie spaces in a workspace).
• Direct use inside agent frameworks such as DSPy, LandGraph, LangChain, and Agent Bricks.
   

Applications using the Conversation APIs orchestrate async workflows. Determine whether a user is starting or continuing a conversation, send the prompt, then poll until generation completes. Responses may be clarifying text or SQL-backed data. When SQL is present, call the results endpoint, then render tables client side. Re-execution helps implement cache layers for repeated questions. Current API support targets Genie chat mode, with agent mode APIs entering private preview.

Multi-tenancy with Genie differs from AI/BI Dashboards today. Genie does not accept an external user ID, so teams typically provision a service principal per tenant and enforce row filters in Unity Catalog against those identities. Visualizations are not returned by the API yet, so applications render their own tables and charts for now. Atlassian illustrates the approach by bringing natural-language data questions directly into Jira workflows.

Make embedding a product capability by standardizing identity and row filters

Treat embedded analytics as a set of architectural choices, not a custom build. Standardize on service principals, OAuth, and Unity Catalog row-level security, then reuse a single AI/BI Dashboard or Genie configuration across tenants. Teams gain faster launches, safer governance, and a clear path to scale.