Agent Bricks: The Agent Developer Platform | Overview

Agent Bricks is the Databricks developer platform for building, deploying, and running custom AI agents at scale, with any framework, model, or cloud. Every era of computing gave developers better tools. Today we're in the era of agents, and Agent Bricks gives you the infrastructure to ship them: • Build: develop your agent in any coding harness, framework, or cloud • Deploy: host and scale on the Databricks agent runtime, with durable execution and automated recovery • Models: use frontier and open-weight models through one unified API with Unity Gateway • Execute: run agent code in isolated, secure sandboxes • Context: Managed agent memory on Lakebase, plus out-of-the-box MCP servers for Genie, web search, Slack, GitHub, and more • Observe: trace and evaluate agents in real time with MLflow, powered by Lakehouse//RT ⏰ TIMESTAMPS: 00:00 – Every era of computing 00:35 – Yesterday's tools get in the way 00:52 – Introducing Agent Bricks: build with any harness, framework, or cloud 01:09 – Deploy with Agent Runtime, models via Unity Gateway 01:28 – Sandboxes, memory, and MCP servers 02:00 – Tracing, evaluation, and getting started 02:31 – Don't build your own agent platform 🔗 LEARN MORE Quickstart: https://docs.databricks.com/aws/en/agents/quickstart Overview: https://docs.databricks.com/aws/en/agents/concepts/ FAQs: https://docs.databricks.com/aws/en/agents/concepts/faq ❓ FAQ What is Agent Bricks? Agent Bricks is the Databricks developer platform for building, deploying, and running custom AI agents at scale. It provides the infrastructure agents need, including a managed runtime, model access through Unity Gateway, sandboxes, memory, MCP tools, and tracing and evaluation, so teams can focus on the agent itself. Which frameworks and models can I use with Agent Bricks? You can develop your agent in any coding harness and framework (for example, LangGraph) and use frontier or open-weight models through one unified API with Unity Gateway, which lets you swap, upgrade, or route between models. How do I deploy an agent with Agent Bricks? Use the Agent Bricks CLI: agentbricks init scaffolds a project, agentbricks dev runs it locally, and agentbricks deploy deploys it to the Databricks agent runtime. The CLI is currently in Beta. How do agents get context and tools on Databricks? Managed agent memory (Beta), built on Lakebase, gives agents durable long-term memory, and out-of-the-box MCP servers connect agents to business context and tools, including Genie, web search, Slack, and GitHub. How can I trace and evaluate my agents? Agent Bricks uses MLflow tracing, on by default, to trace each agent run end to end, so you can evaluate agent quality in real time. Where does my agent run code safely? Agent sandboxes give agents isolated, secure environments to run code, with compute kept separate for each environment. 👉 Start building: https://docs.databricks.com/aws/en/agents/quickstart #Databricks #AgentBricks #AIAgents