datarobot-app-framework

name	datarobot-app-framework
description	Build and deploy applications on DataRobot using the App Framework component system — FastAPI apps, LLM integrations, and agentic workflows (CrewAI, LangGraph, LlamaIndex).

DataRobot App Framework

Build and deploy applications on DataRobot using composable af-component-* building blocks.

Trigger Conditions

Use this skill when the user:

• Wants to build or deploy any application on DataRobot using the App Framework
• Wants to create an agentic workflow (CrewAI, LangGraph, LlamaIndex)
• Wants a FastAPI backend or React frontend deployed to DataRobot
• Wants to integrate an LLM into a DataRobot app or notebook
• Wants to set up an App Framework recipe with any component
• Mentions dr component add, af-component, copier copy, or recipe- in an AF context
• Wants to wire components together (FastAPI ↔ Agent, LLM ↔ Agent, etc.)
• Wants to add tools, auth, or chat history to a DataRobot app

Not This Skill

If the user is asking about any of the following, use datarobot-oss/datarobot-agent-skills instead:

• Training or evaluating ML models (AutoML, tabular data)
• Managing model deployments for predictions
• Feature engineering or feature importance
• Model monitoring, data drift, or model health
• Batch or real-time predictions on tabular data
• SHAP values or model explainability

Note on overlapping vocabulary: The words "agent," "deployment," and "DataRobot" appear in both skill packs but mean different things. In this skill, "agent" means an LLM-based agentic app (CrewAI/LangGraph/LlamaIndex) and "deployment" means a CustomApplication. If the user is talking about ML models or tabular data, route to the OSS pack.

Pick Your Scenario

What you want to build	Scenario
Agent API + playground UI (no custom frontend)	Default — see below
Simple FastAPI app / custom web UI	scenarios/fastapi-app.md
LLM integration in a notebook	scenarios/llm-notebook.md

Component Map

Component	Role
af-component-base	Foundation — always first
af-component-llm	LLM connectivity
af-component-agent	Agent orchestration: CrewAI / LangGraph / LlamaIndex
af-component-fastapi-backend	Simple FastAPI app
af-component-react	React frontend

For extended scenarios, load the relevant file from scenarios/.

Prerequisites

pip install datarobot-cli
curl -LsSf https://astral.sh/uv/install.sh | sh
dr auth set-url && dr auth login

---

Default Recipe: Minimal Agent

Outputs: API endpoint + DataRobot agent playground UI.

Step 1 — Create recipe directory

mkdir recipe-my-agent && cd recipe-my-agent

Convention: always prefix with recipe-. For team projects, create the repo in the DataRobot GitHub org first and clone it.

Step 2 — Scaffold base

uvx copier copy https://github.com/datarobot/af-component-base .

Answer the interactive questions about your recipe name and settings. Defaults are safe.

Step 3 — Add LLM

uvx copier copy https://github.com/datarobot-community/af-component-llm .

Key prompts:

• LLM folder name: llm (default)
• Model name: azure-openai-gpt-4o-mini or as required
• Base answers file: .datarobot/answers/base.yml

Creates infra/infra/llm.py and gateway config files.

Step 4 — Add agent

dr component add agent

Key prompts:

• Agent folder name: agent (default)
• Low-code YAML (NeMo Toolkit)?: No (unless user specifically requests it)
• Framework: CrewAI / LangGraph / LlamaIndex — choose based on user need
• Base answers file: .datarobot/answers/base.yml
• LLM answers file: .datarobot/answers/llm-llm.yml
• MCP answers file: .datarobot/answers/drmcp-mcp_server.yml (optional — skip if not using MCP tools)

Creates:

agent/
├── agent/myagent.py   ← multi-agent workflow (customize here)
├── cli.py             ← local testing tool
├── dev.py             ← local dev server
└── tests/
infra/infra/agent.py   ← Pulumi deployment config

Step 5 — Configure environment

dr dotenv setup

Key prompts:

• Agent port: 8842 (default)
• DataRobot execution environment: select from available environments
• Execution environment version ID: ID of the environment version to use
• Pulumi passphrase: any value (used for state encryption)
• Use case: can leave blank
• LLM Gateway config: select LLM Gateway with External Model

Press enter to accept defaults for most prompts.

Step 6 — Test locally

cd agent
uv run python cli.py execute --user_prompt "Write a blog post about AI in healthcare"
cat execute_output.json | jq -r '.choices[0].message.content'

For iterative development with auto-reload:

dr run dev                                                              # Terminal 1
cd agent && uv run python cli.py execute --user_prompt "Test prompt"   # Terminal 2

Server runs at http://localhost:8842 and reloads on code changes.

Step 7 — Deploy

dr task deploy

Prompts to create a stack (name it anything), previews LLM + Agent resources, then deploys both. Outputs deployment IDs and URLs.

Check deployment info anytime:

dr task infra:info

What you get

• Default: Planner agent (research + outline) + Writer agent (content creation)
• Sequential workflow with MCP tools support
• Agent playground UI in DataRobot
• API endpoint for integration

Deploy updates

dr run deploy

Tear down

dr task infra:down

Customize

Edit agent/agent/myagent.py to:

• Change agent roles, goals, or task descriptions
• Add more agents to the crew
• Integrate additional MCP tools
• Switch framework (CrewAI → LangGraph → LlamaIndex)

Resources

• DataRobot CLI: https://docs.datarobot.com/en/docs/more-info/app-framework/cli.html
• Agent Templates: https://github.com/datarobot-community/datarobot-agent-templates
• Components: search af-component in datarobot / datarobot-community GitHub orgs