Agentic repositories¶

DataRobot publishes several open-source repositories for building, deploying, and operating agentic workflows on the platform. This page groups them into application templates, agentic components, and coding tools so you can find the right starting point for your workflow.

Application templates¶

Application templates are end-to-end scaffolds you clone or configure with the DataRobot CLI. Each template includes application logic (frontend and backend), operational logic (Pulumi infrastructure), and agent or analytics code ready to customize and deploy.

DataRobot Agentic Starter application template¶

The Agentic Starter template is a ready-to-use application for building and deploying agentic workflows. It combines multi-agent framework support, a FastAPI backend, a React frontend, and an MCP server in one repository. Use it to set up a new agentic application with minimal configuration, run everything locally for development and testing, and deploy to DataRobot with a single command.

Key capabilities:

• Supports LangGraph, CrewAI, LlamaIndex, NeMo Agent Toolkit (NAT), and a generic base template.
• Agent code lives under agent/agent/; dr run dev starts the frontend, backend, agent, and MCP server in parallel.
• dr start clones the template, runs the configuration wizard, and creates a .env file with LLM, OAuth, and deployment settings.
• dr run deploy uses Pulumi to deploy the application, agent, and MCP server to DataRobot.

Prerequisites: DataRobot CLI (dr >= 0.2.55), git, uv, Pulumi, Taskfile, and Node.js. The template supports macOS and Linux only; on Windows, use a DataRobot codespace, WSL, or a dev container.

Get started: See Installation, Quickstart, and Customize agents.

Talk to My Data agent¶

Talk to My Data delivers a talk-to-your-data experience for structured files, spreadsheets, and cloud data sources. Upload data, connect to Snowflake or BigQuery, or access datasets from the DataRobot Data Registry, then ask questions and receive recommended analyses with charts, tables, and generated code.

Key capabilities:

• Local uploads, Snowflake, BigQuery, SAP Datasphere, and remote connections through the DataRobot platform (PostgreSQL, Redshift, Databricks, MySQL).
• Handles datasets from thousands to billions of rows with efficient in-app or platform-backed querying.
• Supports LLM gateway direct mode, deployed models, registered models (such as NVIDIA NIM), and external providers through LiteLLM.
• Add custom analysis functions in core/src/core/tools.py for the Python agent.

Get started: Run dr start to configure and deploy, or clone the repository and run dr dotenv setup followed by task deploy for local development.

Talk to My Docs agent¶

Talk to My Docs is a modular template for building document-centric AI applications with multi-agent orchestration, a modern web frontend, and infrastructure-as-code deployment. Connect to documents across Google Drive, Box, SharePoint, and local storage, then query and summarize content through an agentic workflow.

Key capabilities:

• Coordinates specialized agents for document retrieval, reasoning, and response generation.
• Integrates with cloud storage and local file systems through configurable OAuth applications.
• Supports standard and air-gapped deployment paths with Pulumi-managed infrastructure.
• dr start guides you through configuration, .env creation, and deployment in a DataRobot codespace or local environment.

Get started: Run dr start from a DataRobot codespace or after installing the DataRobot CLI.

Agentic components¶

Agentic components are Copier templates in the DataRobot App Framework. Add them to a project with dr component add or uvx copier copy to generate Pulumi infrastructure, Python modules, and stored answers files under .datarobot/answers/.

Most agentic components require af-component-base first. Apply components in dependency order: base, then llm and mcp (repeatable), then agent.

DataRobot Agentic Framework component¶

The agent component provides utilities and templates for single- or multi-agent workflows using NVIDIA NAT, CrewAI, LangGraph, LlamaIndex, and other frameworks. It ships agentic workflow templates, a CLI harness for local testing, and Pulumi-based infrastructure for end-to-end deployment.

When to use it: Add this component when you need agent scaffolding in an App Framework recipe rather than cloning a full application template.

Framework options: base, crewai, langgraph, llamaindex, or nat.

Apply the component:

dr component add https://github.com/datarobot-community/af-component-agent .

Required dependencies: af-component-base, af-component-llm, and af-component-datarobot-mcp.

Local testing: Run task test for all frameworks, or task test-AGENT_FRAMEWORK for a single framework.

DataRobot Agentic LLM component¶

The LLM component adds governed LLM inference to App Framework projects. It generates Pulumi infrastructure and CLI metadata for connecting to the DataRobot LLM gateway, an existing custom model deployment, or external providers such as Azure OpenAI, AWS Bedrock, and Google Vertex AI.

When to use it: Configure LLM access at the infrastructure level without hard-coding provider details in agent code. Because the component is repeatable, you can apply it multiple times under different names for separate LLM instances.

Configuration strategies:

Apply the component:

dr component add https://github.com/datarobot-community/af-component-llm .

See also Configure LLM providers with metadata.

DataRobot Agentic MCP component¶

The MCP component deploys a Model Context Protocol (MCP) server as a DataRobot custom model application. The server runs on FastMCP and exposes DataRobot predictive tools plus integrations for Google Drive, Jira, Confluence, Microsoft 365, Perplexity, Tavily, and DataRobot documentation.

When to use it: Give LLM agents governed access to DataRobot platform APIs and third-party systems through a standard MCP interface. Apply the component multiple times with different mcp_app_name values to run several MCP backends in one project.

Tool categories: DataRobot catalog, modeling, deployments, predictions, vector databases, use cases, and documentation; plus connector tools for Confluence, Jira, Google Drive, and Microsoft Graph (OAuth required).

Apply the component:

dr component add https://github.com/datarobot-community/af-component-datarobot-mcp .

Run locally: After applying, run uv run python -m MCP_APP_NAME from the generated module path.

See also MCP overview and Add tools to agents.

Agentic coding tools and references¶

These repositories provide the CLI, Python libraries, App Framework documentation, and coding-agent skill packs you use alongside templates and components during development.

DataRobot CLI¶

The DataRobot CLI (dr) is the primary command-line interface for agentic application development. It manages OAuth authentication, clones and configures application templates, adds and updates App Framework components, and runs Taskfile tasks.

Key features:

• Authentication: OAuth login with secure credential storage in ~/.config/datarobot/drconfig.yaml.
• Template management: interactive wizard (dr templates setup) to clone and configure templates.
• Component management: dr component add and dr component update for App Framework Copier components.
• Task execution: dr start, dr run dev, and dr task list integrate with template Taskfiles.
• Self-update: dr self update keeps the CLI current.

Quick start:

dr auth set-url
dr auth login
dr start

See Getting started with the DataRobot CLI and the CLI command reference.

DataRobot GenAI¶

datarobot-genai is a Python library (published on PyPI) for building agents on DataRobot. Application templates and App Framework components use it under the hood for LLM routing, tools, streaming, and serving.

Key capabilities:

• get_llm() entry points for LangGraph, LlamaIndex, CrewAI, and NAT, routed through LiteLLM to the DataRobot LLM gateway, deployments, NIM, or external providers.
• drtools standard library and drmcp for hosting custom FastMCP servers on DataRobot.
• Standard event streams so UIs and the platform render agent runs consistently.
• Planner/writer crews, LangGraph graphs, and LlamaIndex AgentWorkflow handoffs with a shared streaming contract.
• Compose LLMs, tools, agents, and evaluators in the low-code workflow.yaml interface (inspired by NeMo Agentic Toolkit).
• Front-end server with distributed tracing, generation and evaluation endpoints, async generations, and WebSocket communication.

Install extras for your framework:

pip install "datarobot-genai[langgraph]"
pip install "datarobot-genai[dragent]"
pip install "datarobot-genai[drmcp]"

Set DATAROBOT_API_TOKEN and DATAROBOT_ENDPOINT before using DataRobot-backed features.

App Framework¶

App Framework Studio is the tooling hub for applying and updating App Framework components. Use it as the reference for the full component catalog and developer guide when composing custom recipes from individual af-component-* repositories.

Component catalog:

See the App Framework developer guide for setup instructions and local documentation workflows.

The repository also publishes a skill pack for coding assistants that work with App Framework recipes and components.

Install App Framework skills:

npx ai-agent-skills install datarobot-community/app-framework

DataRobot Agentic skills¶

Agentic skills are modular capability packages for coding agents. Each skill bundles a SKILL.md file with instructions, examples, and helper scripts so agents load only what they need for a given task.

Skills cover agent building and deployment (datarobot-agent-assist), tabular ML workflows (training, predictions, deployment, monitoring), data preparation, feature engineering, CI/CD for templates, and external agent observability.

Install all skills:

npx ai-agent-skills install datarobot-oss/datarobot-agent-skills

Supported coding agents include Claude Code, Cursor, Codex, VS Code Copilot, Gemini CLI, and others.

For the full skill list, installation options per agent, and usage examples, see DataRobot agentic skills.

How the repositories relate¶

flowchart TB
    subgraph templates ["Application templates"]
        Starter["Agentic Starter"]
        TTMD["Talk to My Data"]
        TTMDocs["Talk to My Docs"]
    end

    subgraph components ["Agentic components"]
        Base["af-component-base"]
        LLM["af-component-llm"]
        MCP["af-component-datarobot-mcp"]
        Agent["af-component-agent"]
    end

    subgraph tools ["Coding tools and references"]
        CLI["DataRobot CLI (dr)"]
        AF["App Framework Studio"]
        GenAI["datarobot-genai"]
        Skills["datarobot-agent-skills"]
    end

    CLI --> templates
    CLI --> AF
    AF --> Base
    Base --> LLM
    Base --> MCP
    LLM --> Agent
    MCP --> Agent
    Starter --> GenAI
    Agent --> GenAI
    Skills -.-> CLI
    AF -.-> Base

Typical paths:

1. Start from an application template: Run dr start, customize the template for your use case, and deploy with dr run deploy or task deploy.
2. Compose a custom App Framework recipe: Apply base, llm, mcp, and agent components with dr component add, then add backend and frontend components as needed.
3. Enhance your coding agent: Install App Framework skills when working with components, or agentic skills for ML and agent-assist workflows.