Production AI Agents

Build your way with your preferred IDE and frameworks like LangChain, LangGraph or LlamaIndex, then deploy to serverless compute in a single step. Or move faster with no code using managed agent builders, including Supervisor Agent and Knowledge Assistant, for multi-step reasoning and governed data retrieval out of the box. Before going live, validate agent behavior and tool use in AI Playground.

Your agents connect directly to the lakehouse — your governed source of truth. Build RAG pipelines, process documents at scale and connect systems like SharePoint and Google Drive while preserving existing access controls. Give agents persistent memory with Lakebase, stored in your lakehouse.

Agent Bricks natively supports the Model Context Protocol (MCP) — the emerging standard for tool integration — giving agents secure access to APIs, databases and SaaS applications. Authenticate services through Unity Catalog with centrally managed credentials and audit trails. Discover and share MCP tools across your organization through a built-in catalog — with every integration visible, permissioned and auditable.

Agent governance and data governance are defined directly in Unity Catalog. Apply role-based access controls to models, tools and connections with complete lineage from outputs to source data while also enforcing rate limits and fallbacks. Set organization-wide policies for prompt injection prevention, sensitive data detection and content filtering — all within a single control plane.

Deploy agents to serverless compute using Databricks Apps without managing any infrastructure. Serve as REST APIs with automatic scaling or schedule on recurring workflows. Monitor end-to-end automatically: capture every interaction, tool call and model invocation with zero code changes. Continuously evaluate quality with LLM-as-judge, custom metrics and human feedback loops.