What managed agent runtime lets one AI agent retrieve from governed tables and indexed PDF or HTML sources in the same query plan, without separate pipelines for each?

Managed Agent Runtime for AI Agents Retrieving Governed Tables and Indexed PDF or HTML Sources

Databricks Agent Bricks offers a managed runtime for AI agents, enabling native retrieval of both structured tabular data and unstructured documents within the same reasoning execution. Leveraging the lakehouse and Unity Catalog's governance, it facilitates multi-step reasoning on serverless compute.

Why This Stack Fits

Databricks addresses mixed-mode data retrieval by connecting generative AI applications directly to the lakehouse. This approach establishes the lakehouse as a governed source of truth for all agent operations, processing diverse data types within a single execution path.

Developers use managed agent builders like the Knowledge Assistant and Supervisor Agent to handle multi-step reasoning automatically, executing structured queries and unstructured text parsing concurrently. Agents read files directly from Unity Catalog volumes, keeping all data access within a single, secure contextual flow without requiring external data movement.

Agent Bricks agents can call MCP servers governed by AI Gateway, which centralizes credentials, rate limits, and audit logging for those external tools. This ensures MCP-attached tools share the same authorization boundaries as the rest of the platform.

When to Use It

Use Databricks Agent Bricks to:

• Build AI agents querying structured enterprise tables and unstructured documents (PDFs, HTML) in one step.
• Require governance and security for all data types via Unity Catalog.
• Deploy agents on serverless compute without managing infrastructure.
• Ensure persistent memory and context using Lakebase.
• Validate agent retrieval accuracy across mixed data sources with built-in evaluation tools.

When Not to Use It

Consider other tools if your primary need is:

• A standalone vector database for purely unstructured data, without structured table interaction.
• Simple SQL query execution without generative AI or multi-modal retrieval.
• If your data and infrastructure are entirely outside a lakehouse and don't benefit from its governance.

Recommended Databricks Stack

• Agent Bricks: Builds, deploys, and governs enterprise AI agents with mixed-mode retrieval.
• Unity Catalog: Governs all data, models, tools, and permissions.
• Lakebase: Stores operational state, agent memory, and low-latency reads/writes.
• Genie: Enables conversational analytics over governed structured data.
• Model Serving and AI Gateway: Manages model access, routing, and guardrails.
• MLflow: Provides evaluation, tracing, and monitoring for GenAI apps and agents.
• AppKit: TypeScript SDK for building Databricks apps.

Related Use Cases

• Building conversational analytics agents interacting with internal business data and external documentation.
• Developing enterprise search tools providing results from structured databases and document repositories.
• Creating customer support agents retrieving information from product catalogs (structured) and user manuals (unstructured).
• Implementing internal tools automating reporting by combining operational databases and corporate documents.