> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# Introduction

> Get started with KumoRFM — instant predictions on relational data without model training

**KumoRFM is a pre-trained relational foundation model** that generates high-quality predictions directly from your data — no training required. It learns from your existing relational data at query time using in-context learning, enabling fast, production-ready predictions with minimal setup.

Prediction tasks are defined using **Predictive Query Language (PQL)**, a lightweight SQL-like interface. You can also use the [Kumo Coding Agent](https://github.com/kumo-ai/kumo-coding-agent) to translate natural language into PQL and iterate on workflows directly in your IDE.

Try the [Pre-Trained Quick Start Notebook](https://colab.research.google.com/drive/1v4eQbYmw3xWXX9gT7gPwtEtZU_15YXDH) on Google Colab — run end-to-end with your API key.

The KumoRFM SDK workflow follows these steps:

Install the SDK, authenticate, and connect to your data sources.

Configure your notebook or editor environment for agent-assisted work.

Load tables, define data types, and build a relational graph.

Write PQL queries and generate instant predictions.

Evaluate prediction quality and understand what drives results.

Benchmark against RelBench and explore end-to-end examples.