Get started¶

The Agentic Starter template repository provides a ready-to-use application template for building and deploying agentic workflows with multi-agent frameworks, a FastAPI backend server, a React frontend, and an MCP server. The template streamlines the process of setting up new agentic applications with minimal configuration requirements and supports local development and testing, as well as one-command deployments to production environments within DataRobot.

This guide covers installing the prerequisite tools and configuring your environment, then creating, deploying, and testing an agentic application using DataRobot's pre-built templates.

System requirements¶

Ensure your system meets the minimum requirements for running the Agentic Starter template:

• Operating system: macOS, Linux, or Windows
• Python: Version 3.10 or higher
• Memory: At least 4 GB of RAM

Install prerequisite tools¶

Before you begin, you'll need the following tools installed. If you already have these tools installed, ensure that they are at the required version (or newer) indicated in the table below. For example commands to install the tools, see the Detailed installation commands section.

Windows prerequisites¶

Complete these steps on Windows before you clone the repository. Skipping them checks symlinks out as plain text files and leaves the working tree broken.

1. Enable symlink support in Git:

git config --global core.symlinks true

1. Grant permission to create symlinks using one of the following options:

2. Developer Mode (recommended): On Windows 11, open Settings → System → Advanced → Developer Mode and turn Developer Mode on. See Enable your device for development for details.

3. Administrator terminal: Launch PowerShell or Windows Terminal with Run as administrator and run every repo operation from that elevated session. At minimum, use an elevated session for git clone, dr start, dr run deploy, and any git checkout or git pull that touches symlinked paths. This template uses Git symlinks at .claude/skills, fastapi_server/core, infra/infra/llm.py, and infra/infra/oauth.py.

4. After Git is installed, ensure that the following directory is present in your system PATH: [path_of_git_installation]\usr\bin. For example, if Git is installed in C:\Program Files\Git, add C:\Program Files\Git\usr\bin to your PATH using the following commands:

$dir = 'C:\Program Files\Git\usr\bin'  # Change if Git is installed elsewhere.
$p = [Environment]::GetEnvironmentVariable('PATH', 'User')
[Environment]::SetEnvironmentVariable('PATH', "$p;$dir", 'User')

This location provides Linux helper commands required for the Agentic Starter to work correctly. Close and reopen the terminal (or IDE) after running the command so new processes pick up the change.

Detailed installation commands¶

The following sections provide example installation commands for macOS, Linux (Debian/Ubuntu/DataRobot codespace), and Windows (PowerShell). Click the tab below that corresponds to your operating system:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" # If homebrew is not already installed

brew install datarobot-oss/taps/dr-cli uv pulumi/tap/pulumi go-task node git

xcode-select --install

curl https://cli.datarobot.com/install | sh
sudo apt-get update
sudo apt-get install -y python3 python3-pip python3-venv
sudo apt-get install -y build-essential
sudo apt-get install -y git
curl -LsSf https://astral.sh/uv/install.sh | sh
curl -fsSL https://get.pulumi.com | sh
sh -c "$(curl --location https://taskfile.dev/install.sh)" -- -d
sudo apt-get install -y nodejs npm

irm https://cli.datarobot.com/winstall | iex
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
winget install Git.Git
winget install Pulumi.Pulumi
winget install Task.Task
winget install OpenJS.NodeJS
winget install Microsoft.VisualStudio.2022.BuildTools --force --override "--wait --passive --add Microsoft.VisualStudio.Component.VC.Tools.x86.x64 --add Microsoft.VisualStudio.Component.Windows11SDK.22621"

Initialize your application¶

Run the following command to start the local development environment:

dr start

This command starts the DataRobot CLI's interactive wizard (dr start in the CLI command reference). It automatically clones the application repository and creates a .env file in the root directory populated with environment variables you specify. The wizard provides guidance and context for each step; expand the reference below for the complete list of steps.

After dr start completes successfully, verify the following:

• A .env file in your project root.
• Your application directory created (typically named datarobot-agent-application or based on your application name).

Now that your application is configured, proceed to the next section.

Run your agent¶

Navigate to the application directory created during dr start:

cd datarobot-agent-application # or the custom directory name you specified during the wizard, if different

Then, run the following command to start all components of the application:

dr run dev

This starts four processes, running in parallel:

• Application frontend
• Application backend
• Agent
• MCP server

Once all services are running:

1. Open your web browser and navigate to http://localhost:5173.
2. Confirm that the agent application interface appears.
3. Try sending a test message to verify everything is working.

From here, start customizing your agent by adding your own logic and functionality. See the Develop your agent section for more details.

Develop your agent¶

Now that your agent has been built and tested, you are ready to customize it by adding your own logic and functionality. See the following documentation for more details:

• Customize your agent
• Add tools to your agent
• Configure LLM providers
• Add Python requirements
• Manage prompts

Deploy your agent¶

Next, deploy your agent to DataRobot, which requires a Pulumi login.

Run the following command to deploy your agent:

dr task run deploy

For more on the task and run commands, see the CLI task command and CLI run command.

Once deployment is complete, the script displays the deployment details, as shown in the example below. The deployment details vary based on your configuration.

Outputs:
    AGENT_DEPLOYMENT_ID                               : "69331fad5e07469e7c4f5c6f"
    Agent Custom Model Chat Endpoint [apptest] [agent]: "https://datarobot.com/api/v2/genai/agents/fromCustomModel/69331f816e1bf9f1890d5d1d/chat/"
    Agent Deployment Chat Endpoint [apptest] [agent]  : "https://datarobot.com/api/v2/deployments/69331fad5e07469e7c4f5c6f/chat/completions"
    Agent Execution Environment ID [apptest] [agent]  : "680fe4949604e9eba46b1775"
    Agent Playground URL [apptest] [agent]            : "https://datarobot.com/usecases/69331e4c3be0efe3b95a7be0/agentic-playgrounds/69331e4d1c036307186c9b16/comparison/chats"
    Agentic Starter [apptest]             : "https://datarobot.com/custom_applications/6933204a9e21e9b59b5a7bee/"
    DATABASE_URI                                      : "sqlite+aiosqlite:////tmp/agent_app/.data/agent_app.db"
    DATAROBOT_APPLICATION_ID                          : "6933204a9e21e9b59b5a7bee"
    DATAROBOT_OAUTH_PROVIDERS                         : (json) []

    LLM_DEFAULT_MODEL                                 : "azure/gpt-4o-2024-11-20"
    SESSION_SECRET_KEY                                : "secretkey123"
    USE_DATAROBOT_LLM_GATEWAY                         : "1"
    [apptest] [mcp_server] Custom Model Id            : "69331eebb49131d3d5430ac7"
    [apptest] [mcp_server] Deployment Id              : "69331f1f30548f83b668d9dc"
    [apptest] [mcp_server] MCP Server Base Endpoint   : "https://datarobot.com/api/v2/deployments/69331f1f30548f83b668d9dc/directAccess/"
    [apptest] [mcp_server] MCP Server MCP Endpoint    : "https://datarobot.com/api/v2/deployments/69331f1f30548f83b668d9dc/directAccess/mcp"

Restricted network setup¶

Complete this section only if you're installing, initializing, or deploying from an environment with no direct internet access (for example, an air-gapped environment); otherwise, skip it.

Configure Pulumi to install the DataRobot plugin from an internal proxy instead of GitHub. Setting these environment variables redirects all Pulumi plugin downloads to your internal proxy and disables external update checks.

Add the following variables to your .env file:

PULUMI_SKIP_UPDATE_CHECK=1
PULUMI_DATAROBOT_DEFAULT_URL=http://internal-proxy-for-pulumi
# OPTIONAL
PULUMI_DATAROBOT_PLUGIN_VERSION=v0.10.27

Python packages¶

In restricted network environments, uv sync operations fail when attempting to reach the public PyPI. To resolve this, configure uv to use your internal PyPI proxy.

To configure the proxy, edit the agent/pyproject.toml file in your agent project and uncomment the [tool.uv.pip] section, replacing the URL with your internal PyPI proxy. For example:

[tool.uv.pip]
extra-index-url = ["https://your-internal-pypi-proxy.example.com/simple/"]