DataRobot - OSS

Homepage · Documentation · App Framework Docs · Support

Welcome to DataRobot's open source home! 👋

Here you'll find open-source software from our Research & Development and Customer-Facing teams — the tools, scripts, and libraries we use to build and work with DataRobot ourselves, shared back with the community.

Disclaimer: these are first-party open-source repositories, provided as-is with no official support.

DataRobot's code is spread across four GitHub organizations. Which one you want depends on what you're doing:

Agentic application work in particular spans datarobot-oss (runtime libraries and the CLI) and datarobot-community (the components and templates you compose) — the next section maps out how the pieces fit together.

The App Framework (docs: af.datarobot.com) is a code-first way to build an agentic application and ship it to DataRobot. Rather than one monolithic template, you compose small copier components — the af-component-* repos — into a project, then deploy it with a single task.

flowchart LR
  DEV["👩‍💻 developer<br/>(+ coding agent)"] --> CLI["dr CLI"]
  CLI -->|"uvx copier"| COMP["af-component-*"]
  COMP --> APP["agent application<br/>agent/ · mcp_server/<br/>fastapi_server/ · frontend_web/ · infra/"]
  APP -.->|"builds on"| GENAI["datarobot-genai"]
  APP -->|"task deploy → Pulumi"| PF["DataRobot platform<br/>agent deployment · MCP deployment<br/>custom application"]
  PF --> USER["💬 end users"]

How the flow works

1. Scaffold. A developer — often working alongside a coding agent loaded with datarobot-agent-skills — runs the dr CLI, which composes af-component-* modules with copier.
2. Compose. Components render into a project such as datarobot-agent-application, producing an agent/ package plus whatever backend, frontend, MCP server, and Pulumi infrastructure you asked for.
3. Build. The generated agent runs on the datarobot-genai runtime library, routes model calls through the DataRobot LLM Gateway (or an LLM deployment / NIM of your choosing), and pulls tools from the MCP server it was built with.
4. Deploy. dr task deploy runs Pulumi — datarobot-pulumi-utils over the pulumi-datarobot provider — which provisions the agent as a custom model and serverless deployment, plus the MCP server and the chat application.
5. Use. End users talk to the deployed application, which streams the conversation to and from the agent deployment.

Each component is a copier template with declared dependencies, so base is always applied first and the rest layer on top. Components marked repeatable can be applied more than once (two MCP servers, three LLMs, and so on).

Jump to a category:

• 🤖 AI Agents & Coding Skills
• 🧠 GenAI & Model Serving
• ⌨️ CLI & Developer Tools
• 🏗️ Install & Infrastructure
• 🔁 CI/CD & Automation
• 🚀 Application Templates & Samples
• 🔌 Integrations

Bring DataRobot into your AI coding agents, and build agentic systems you can trust.

Provision and deploy the infrastructure DataRobot runs on.

Reusable automation for building, deploying, and governing DataRobot workloads.

Ready-to-use starters for building custom applications on DataRobot. For agentic applications, start with the Agent Framework instead.

• 💬 Join the conversation in #all-datarobot-community on Slack.
• 🐛 Found a bug or have an idea? Open an issue on the repository in question.
• 📚 Product documentation lives at docs.datarobot.com; App Framework documentation at af.datarobot.com.