Skip to content

Template system structure

This page provides an understanding of how DataRobot organizes and configures application templates.

Overview

DataRobot templates are Git repositories that contain application code, configuration, and metadata to deploy custom applications to DataRobot. The CLI provides tools to clone, configure, and manage these templates.

Template repository structure

A typical template repository:

my-datarobot-template/
├── .datarobot/              # Template metadata
│   ├── prompts.yaml         # Configuration prompts
│   ├── config.yaml          # Template settings
│   └── cli/                 # CLI-specific files
│       └── bin/             # Quickstart scripts
│           └── quickstart.sh
├── .env.template            # Environment variable template
├── .taskfile-data.yaml      # Taskfile configuration (optional)
├── .gitignore
├── README.md
├── Taskfile.gen.yaml        # Generated task definitions
├── src/                     # Application source code
│   ├── app/
│   │   └── main.py
│   └── tests/
├── requirements.txt         # Python dependencies
└── package.json             # Node dependencies (if applicable)

Template metadata

.datarobot directory

The .datarobot directory contains template-specific configuration:

.datarobot/
├── prompts.yaml        # User prompts for setup wizard
├── config.yaml         # Template metadata
└── README.md          # Template-specific docs

prompts.yaml

Defines interactive configuration prompts. See Interactive configuration for more details.

Review the example prompt yaml configuration below.

prompts:
  - key: "app_name"
    env: "APP_NAME"
    help: "Enter your application name"
    default: "my-app"
    optional: false

  - key: "deployment_target"
    env: "DEPLOYMENT_TARGET"
    help: "Select deployment target"
    options:
      - name: "Development"
        value: "dev"
      - name: "Production"
        value: "prod"

config.yaml

Template metadata and settings:

name: "My DataRobot Template"
version: "1.0.0"
description: "A sample DataRobot application template"
author: "DataRobot"
repository: "https://github.com/datarobot/template-example"

# Minimum CLI version required
min_cli_version: "0.1.0"

# Tags for discovery
tags:
  - python
  - streamlit
  - machine-learning

# Required DataRobot features
requirements:
  features:
    - custom_applications
  permissions:
    - CREATE_CUSTOM_APPLICATION

Environment configuration

.env.template

Review a template for environment variables. Note that the commented lines are optional.

# Required configuration
APP_NAME=
DATAROBOT_ENDPOINT=

# Optional configuration (commented out by default)
# DEBUG=false
# LOG_LEVEL=info

# Database configuration (conditional)
# DATABASE_URL=postgresql://localhost:5432/mydb
# DATABASE_POOL_SIZE=10

# Authentication
# AUTH_ENABLED=false
# AUTH_PROVIDER=oauth2

.env (Generated)

Created by the CLI during setup, the .env file contains actual values.

[!WARNING] Add the .env file to .gitignore to ensure it is never committed.

# Required configuration
APP_NAME=my-awesome-app
DATAROBOT_ENDPOINT=https://app.datarobot.com

# Optional configuration
DEBUG=true
LOG_LEVEL=debug

# Database configuration
DATABASE_URL=postgresql://localhost:5432/mydb
DATABASE_POOL_SIZE=5

Quickstart scripts

Templates can optionally provide quickstart scripts to automate application initialization. These scripts are executed by the dr start command.

Quickstart scripts must be placed in .datarobot/cli/bin/.

Naming conventions

Scripts must start with quickstart (case-sensitive):

  • ✅ quickstart
  • ✅ quickstart.sh
  • ✅ quickstart.py
  • ✅ quickstart-dev
  • ❌ Quickstart.sh (wrong casing)
  • ❌ start.sh (wrong name)

If there are multiple scripts matching the pattern, the first one found in lexicographical order will be executed.

Platform requirements

Review the requirements for different platforms below.

Unix/Linux/macOS

  • Must have executable permissions (chmod +x)
  • Can be any executable file (shell script, Python script, compiled binary, etc.)

Windows

  • Must have an executable extension: .exe, .bat, .cmd, or .ps1

When to use quickstart scripts

Quickstart scripts are useful for:

  • Multi-step initialization: When your application requires several setup steps
  • Dependency management: Install packages or tools before starting
  • Environment validation: Check prerequisites before launch
  • Custom workflows: Template-specific initialization logic

Fallback behavior

If dr start does not find a quickstart, it automatically launches the interactive dr templates setup wizard instead to ensure that you can always get started even without a custom script.

Task definitions

Taskfile.gen.yaml / Taskfile.yaml

The CLI generates a root Taskfile to aggregate component tasks.

  • Taskfile.gen.yaml: Automatically generated and used by dr run / dr task run.
  • Taskfile.yaml: Generated by dr task compose for use with the native task command or for manual inspection.

These files include a dotenv directive to load environment variables from .env.

[!WARNING] Component taskfiles cannot have their own dotenv directives. The CLI detects conflicts and prevents generation if a component taskfile already has a dotenv declaration.

The generated structure is shown below.

version: '3'

dotenv: [".env"]

includes:
  backend:
    taskfile: ./backend/Taskfile.yaml
    dir: ./backend
  frontend:
    taskfile: ./frontend/Taskfile.yaml
    dir: ./frontend

Component taskfiles

Component directories define their own tasks:

Review the structure of backend/Taskfile.yaml below.

version: '3'

# Note: No dotenv directive are allowed here

tasks:
  dev:
    desc: Start development server
    cmds:
      - python -m uvicorn src.app.main:app --reload

  test:
    desc: Run tests
    cmds:
      - pytest src/tests/

  build:
    desc: Build application
    cmds:
      - docker build -t {{.APP_NAME}} .

Running tasks

The dr run command requires a .env file to be present:

# List all available tasks
dr run --list

# Run a specific task
dr run dev

# Run multiple tasks
dr run lint test

# Run tasks in parallel
dr run lint test --parallel

If you're not in a DataRobot template directory (no .env file), you'll see the following message:

You don't seem to be in a DataRobot Template directory.
This command requires a .env file to be present.

Taskfile configuration data

Template authors can optionally provide a .taskfile-data.yaml file to configure the generated Taskfile. This file allows specifying port numbers for development servers and other configuration data.

See dr task compose documentation for complete details on the file format and usage.

Multi-level configuration

Templates can have nested .datarobot directories for component-specific configuration:

my-template/
├── .datarobot/
│   └── prompts.yaml          # Root level prompts
├── backend/
│   ├── .datarobot/
│   │   └── prompts.yaml      # Backend prompts
│   └── src/
├── frontend/
│   ├── .datarobot/
│   │   └── prompts.yaml      # Frontend prompts
│   └── src/
└── .env.template

Discovery order

The CLI discovers prompts in this order:

  1. Root .datarobot/prompts.yaml
  2. Subdirectory prompts (depth-first search, up to depth 2)
  3. Merged and deduplicated

Example: backend prompts

backend/.datarobot/prompts.yaml:

backend:
  - key: "api_port"
    env: "API_PORT"
    help: "Backend API port"
    default: "8000"

  - key: "database_url"
    env: "DATABASE_URL"
    help: "Database connection string"

Example: frontend prompts

frontend/.datarobot/prompts.yaml:

frontend:
  - key: "ui_port"
    env: "UI_PORT"
    help: "Frontend UI port"
    default: "3000"

  - key: "api_endpoint"
    env: "API_ENDPOINT"
    help: "Backend API endpoint"
    default: "http://localhost:8000"

Template lifecycle

1. Discovery

Templates are discovered from DataRobot:

# List available templates
dr templates list

Output: 

Available templates:
* python-streamlit     - Streamlit application template
* react-frontend       - React frontend template
* fastapi-backend      - FastAPI backend template

2. Cloning

Clone a template to your local machine:

# Set up a template (clones and configures interactively)
dr templates setup

This: - Clones the Git repository - Sets up directory structure - Initializes configuration files

3. Configuration

Configure the template interactively:

# Full setup wizard
dr templates setup

# Or configure existing template
cd my-template
dr dotenv setup

4. Development

Work on your application:

# Run development server (requires .env file)
dr run dev

# Run tests
dr run test

# Build for deployment
dr run build

[!NOTE] All dr run commands require a .env file in the current directory. If you see an error about not being in a template directory, run dr dotenv setup to create your .env file.

5. Deployment

Deploy to DataRobot:

dr run deploy

Template types

Python templates

python-template/
├── .datarobot/
├── requirements.txt
├── setup.py
├── src/
│   └── app/
│       └── main.py
├── tests/
└── .env.template

Key features

  • Python dependencies in requirements.txt
  • Source code in src/
  • Tests in tests/

Node.js templates

node-template/
├── .datarobot/
├── package.json
├── src/
│   └── index.js
├── tests/
└── .env.template

Key features

  • Node dependencies in package.json
  • Source code in src/
  • npm scripts integration

Multi-language templates

full-stack-template/
├── .datarobot/
├── backend/
│   ├── .datarobot/
│   ├── requirements.txt
│   └── src/
├── frontend/
│   ├── .datarobot/
│   ├── package.json
│   └── src/
├── docker-compose.yml
└── .env.template

Key features

  • Separate backend and frontend
  • Component-specific configuration
  • Docker composition

Best practices

Version control

[!WARNING] Always exclude .env and Taskfile.gen.yaml from version control by adding them to .gitignore. The CLI generates Taskfile.gen.yaml automatically.

# .gitignore should include:
.env
Taskfile.gen.yaml
*.log
__pycache__/
node_modules/
dist/

Documentation

Include a clear README.

# My template

## Quick start

1. Set up: `dr templates setup`
2. Run: `dr run dev`

## Available tasks

- `dr run dev`: development server.
- `dr run test`: run tests.
- `dr run build`: build for production.

Sensible defaults

Provide defaults in .env.template.

# Good defaults for local development
API_PORT=8000
DEBUG=true
LOG_LEVEL=info

Clear prompts

Use descriptive help text.

prompts:
  - key: "database_url"
    help: "PostgreSQL connection string (format: postgresql://user:pass@host:5432/dbname)"

5. Organized structure

Keep related files together.

src/
├── api/          # API endpoints
├── models/       # Data models
├── services/     # Business logic
└── utils/        # Utilities

Template updates

Updating templates

# Update to latest version
git pull origin main

# Re-run configuration if needed
dr dotenv setup

Creating your own template

1. Start with base structure

mkdir my-new-template
cd my-new-template
git init

2. Add template files

Create the necessary files:

# Configuration
mkdir .datarobot
touch .datarobot/prompts.yaml
touch .env.template

# Application structure
mkdir -p src/app
touch src/app/main.py

# Tasks
touch Taskfile.gen.yaml

3. Define prompts

.datarobot/prompts.yaml:

prompts:
  - key: "app_name"
    env: "APP_NAME"
    help: "Enter your application name"
    optional: false

4. Create an environment template

.env.template:

APP_NAME=
DATAROBOT_ENDPOINT=

5. Define tasks

Create component Taskfiles (e.g., backend/Taskfile.yaml):

version: '3'

tasks:
  dev:
    desc: Start development server
    cmds:
      - echo "Starting {{.APP_NAME}}"

6. Configure Taskfile data

Optional. Create .taskfile-data.yaml to provide additional configuration for the generated root taskfile:

# .taskfile-data.yaml
# Optional configuration for dr task compose

# Ports to display when running dev task
ports:
  - name: Backend
    port: 8080
  - name: Frontend
    port: 5173

This allows developers using your template to see which ports services run on when they execute task dev.

7. Test the template

# Test the setup locally
dr templates setup

# Verify configuration
dr run --list

8. Publish the template

# Push to GitHub
git add .
git commit -m "Initial template"
git push origin main

# Register with DataRobot (contact your admin)

See also