> 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.

# Quickstart: Kumo Coding Agent

> Use coding agents like Claude Code and Codex to iterate on KumoRFM predictions

#### [Kumo Coding Agent](https://github.com/kumo-ai/kumo-coding-agent)

Install the agent, context files, and skills from GitHub to get started.

This guide helps you get started with the [Kumo Coding Agent](https://github.com/kumo-ai/kumo-coding-agent) for KumoRFM Pre-Trained workflows.

Use this page when you already have a project, notebook, or editor environment available and want a short path to productive agent-assisted work.

What this quick start covers:

* choosing an environment for agent-assisted work
* connecting the Kumo Coding Agent to your local project or notebook flow
* using Codex or Claude Code for setup help, graph definition, and predictive query authoring
* verifying that the agent can work with your KumoSDK project context

Choose your environment:

#### [Jupyter in VS Code](/rfm/setup/jupyter-vscode)

Notebook workflow in VS Code

#### [Jupyter in PyCharm](/rfm/setup/jupyter-pycharm)

Notebook workflow in PyCharm

#### [VS Code](/rfm/setup/vscode-kumo-agent)

Script and project workflow

#### [Cursor](/rfm/setup/cursor)

AI-native editor

#### [Codex](/rfm/setup/codex)

Desktop app or CLI

#### [Claude Code](/rfm/setup/claude-code)

Desktop app or CLI

## Try These Prompts

All examples use a sample e-commerce dataset: `s3://kumo-sdk-public/rfm-datasets/online-shopping`

Copy and paste these directly into your coding agent (Claude Code, Codex, or Cursor).

### Explore the Data

Start here. The agent will load the dataset, inspect every table, and summarize what it finds.

```text
Load the data from s3://kumo-sdk-public/rfm-datasets/online-shopping and
tell me what tables are available, how many rows each has, and what the
columns and types look like.
```

### Build a Graph

Once you understand the data, ask the agent to build a relational graph.

```text
Build a KumoRFM graph from the online-shopping dataset at
s3://kumo-sdk-public/rfm-datasets/online-shopping. Inspect the schema,
infer relationships between tables, validate the graph, and visualize it.
```

### Write a Prediction Query

Ask the agent to write a PQL query for a specific business question.

```text
Using the online-shopping graph, write a PQL query that predicts whether
each customer will make a purchase in the next 14 days.
```

### Run an End-to-End Prediction

This is the full workflow in one prompt. The agent will load data, build the graph, write PQL, run the prediction, and show results.

```text
Run a complete KumoRFM prediction on the online-shopping dataset at
s3://kumo-sdk-public/rfm-datasets/online-shopping. Predict the total
spend per customer over the next 30 days. Show me the results and
evaluate the prediction quality.
```

### Explain the Results

After running a prediction, ask the agent to explain what drove the results.

```text
Pick one customer from the predictions and explain what features were
most important in driving their predicted spend.
```

### Try a Different Question

Change the prediction task to explore what else the data can answer.

```text
Using the same online-shopping graph, predict which product category
each customer is most likely to buy from next.
```

---

## What to Expect

When you run these prompts, the agent will:

1. **Inspect the data first** before writing any code
2. **Build the graph** with correct table names and relationships
3. **Write PQL** using real column names from the schema
4. **Run the prediction** and show sample output
5. **Evaluate** the results with appropriate metrics

If something goes wrong, use the agent to debug and iterate until the prediction runs successfully:

```text
The prediction failed. Can you look at the error, fix the issue, and
re-run it?
```

```text
The evaluation metrics look weak. Can you try a different time window
or aggregation and compare the results?
```

```text
Can you try a different prediction target on this dataset and see if
we get better results?
```

The goal is to keep iterating with the agent until you have a working, high-quality prediction notebook or script.

---

## Next Steps

* [Kumo Coding Agent on GitHub](https://github.com/kumo-ai/kumo-coding-agent) for agent source, context files, and skills
* [Setup](/rfm/sdk-getting-started) for SDK fundamentals
* [Make Predictions](/rfm/make-predictions) for PQL query reference
* [Prediction Types](/rfm/prediction-types) for supported prediction types
* [Environment Setup](/rfm/setup/jupyter-vscode) for detailed editor and notebook setup