DataRobot App Framework
The App Framework is the software and tooling that makes building, updating, and shipping DataRobot applications fast. It gives developers and data scientists a code-first starting point, similar to Helm charts or full-stack starters, but built specifically for DataRobot.
What it is
App templates are code-first recipes that bundle everything needed to build and run a full-stack DataRobot application: infrastructure-as-code, a local development loop, CI/CD, and deployment to DataRobot Custom Applications, all from a single repository. Foundation templates already exist for:
- Talk to My Docs — Guarded RAG Assistant.
- Talk to My Data — Data Analyst.
- DataRobot Agentic starter — Multi-agent orchestration starter.
- DataRobot MCP — FastMCP server with DataRobot predictive tools and third-party integrations.
This documentation explains how the framework is organized, how components fit together, and how to build or extend an application template. For new readers, start here and continue to the getting-started and design pages.
Prerequisites
Most guides assume the following:
- A DataRobot account and API token.
- The DataRobot CLI installed.
uvinstalled.- A GitHub repository for creating or cloning an application template.
How it works
The framework is built around three ideas:
Components are copier-style templates, each a top-level folder or file set, layered into a recipe repository to add capabilities. Apply them one at a time and answer a few questions. Answers are recorded in .datarobot/, so the stack can be updated automatically later.
| Component | What it adds |
|---|---|
| base | Task runner, Pulumi project, CI/CD scaffolding, .datarobot/ configuration. |
| fastapi-backend | FastAPI server, local development loop, Custom Application deployment. |
| react | React + Vite frontend with a development proxy and static asset build. |
| llm | LLM Gateway or external model integration. |
| agent | CrewAI, LangGraph, LlamaIndex, or NeMo agent workflow. |
| datarobot-mcp | FastMCP server with DataRobot predictive tools and third-party integrations. |
The DataRobot CLI (dr) is the primary interface. It handles authentication, environment setup, local development, and deployment, so manual YAML editing and command lookup are not required.
The Declarative API (via Pulumi) manages DataRobot resources, including use cases, deployments, playgrounds, and custom applications, as infrastructure-as-code. This makes the stack reproducible across environments.
For readers who are new to the terminology:
- A recipe is the repository for one application template or application instance.
- A component is a reusable template that adds one capability, such as a FastAPI backend or React frontend.
- An answers file is the YAML file in
.datarobot/answers/that stores the choices made when applying a component.
Get started
- 0-Vibe: Build your first application — Go from an empty repository to a deployed application in minutes.
- Design — Understand the architecture, repository model, and building blocks.
- Components — Learn what each component adds and the order in which to apply them.
- Adding custom pages — Extend FastAPI templates with new routes and templates.
- Adding a vector database — Ground your agent in real documents.
- Using dr-ui — Install shared React components from the dr-ui registry.
- Developer guide — Run the docs site locally and contribute.
- Skills — AI agent skills for your coding assistant.