> 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.

# Overview

> Introduction to KumoRFM — the relational foundation model for predictive AI on relational data

[Kumo](https://kumo.ai/) is a next-generation predictive AI platform for relational data that lets you generate high-quality predictions directly from your data warehouse without model training or feature engineering.

At its core is [KumoRFM](https://kumo.ai/research/kumo_relational_foundation_model.pdf), a **relational foundation model** trained on large-scale real and synthetic relational data that performs in-context predictions. It learns from your existing data at query time without training, enabling fast, production-ready predictions with minimal setup. For use cases that require additional optimization, KumoRFM can also be **fine-tuned** for task-specific performance.

Prediction tasks are defined using **Predictive Query Language (PQL)**, a lightweight SQL-like interface. **Kumo's coding agent** translates natural language into PQL and helps you author and iterate on SDK-based workflows directly in your notebook or IDE of choice.

The KumoRFM SDK supports two modes:

For instant, in-context predictions. No model training required.

For optimizing performance on specific tasks with custom training.

AI agent that translates natural language into PQL and SDK workflows.

## Get Started

Open in Google Colab

Open in Google Colab