Agent Bricks quickstart

Experimental

The Agent Bricks CLI is experimental. Commands and behavior can change.

This quickstart takes you from an empty directory to a deployed custom agent with the Agent Bricks CLI (agentbricks). The CLI scaffolds the agent from a framework template and sets up model access, memory, sessions, and tracing for you, so you can focus on the agent's logic.

Prerequisites

The Databricks CLI, installed.
Python 3.10 or above, and uv.
Install the Agent Bricks CLI:
```
pip install databricks-agentbricks
```

Step 1: Authenticate

To authenticate to your workspace with OAuth, save a named profile, and set it as the Agent Bricks CLI's default, run the following:

databricks auth login --host https://<your-workspace-url> --profile <profile>
agentbricks login --profile <profile>

Step 2: Scaffold the project

To scaffold a LangGraph agent with a browser chat app, run the following:

agentbricks init --framework langgraph my-agent
cd my-agent

agentbricks init writes the agent code and agent.toml, which declares the memory store, session store, and tracing experiment the deployed agent uses. Pass --framework openai to scaffold an OpenAI Agents SDK project instead.

Step 3: Run the agent locally

To run the agent on your machine, run the following:

agentbricks dev

Open http://localhost:8000 to chat with the agent. The chat UI links to a local MLflow server, where you can view a trace of each request.

Step 4: Deploy the agent

To deploy the agent to the agent runtime, run the following:

agentbricks deploy my-agent

The CLI creates the stores and tracing experiment that agent.toml declares, grants the agent access to them, and deploys it as a Databricks app named agent-bricks-my-agent. When it finishes, it prints the app's URL. Open the URL to chat with your deployed agent.

Where to next

Agent Bricks CLI for each step in detail, querying the agent from your terminal or code, and bringing an existing agent.
Deploy agents on the agent runtime for how the deployed agent runs, recovers, and authenticates.
Agent memory and sessions to work with the agent's conversation history and long-term memory.