VS Code + Kumo Agent

Overview

This guide walks you through setting up the KumoRFM SDK and Kumo Coding Agent for use with KumoRFM in a VS Code project workflow. Unlike the notebook-based setup, this guide focuses on working with regular Python files, the VS Code integrated terminal, and coding-agent workflows that edit scripts, configuration, and project files directly. This guide assumes minimal prior experience with VS Code, Python virtual environments, and coding agents. If you already have a preferred local setup, you can jump directly to the relevant sections.

Prerequisites

Make sure you have:

VS Code installed (Download)
Python installed at the system level
Access to a KumoRFM environment, including an API key and any required API URL
An OpenAI or Anthropic subscription (recommended if you plan to use the Kumo Coding Agent)
- OpenAI Codex (Subscribe)
- Anthropic Claude Code (Subscribe)
Homebrew (macOS package manager) (Install)
GitHub CLI installed and authenticated (required for installing coding-agent skills and some GitHub-powered workflows)

🖥 Terminal

brew install gh
gh auth status
gh auth login

Part 1: Setup for VS Code + KumoRFM SDK

Step 1: Create or Open a Project Folder

Start with a normal VS Code project folder rather than a notebook file.

Open VS Code
Choose File -> Open Folder
Create or open a project directory such as kumo-demo

A simple starter structure works well:

kumo-demo/
├── .venv/
├── data/
├── scripts/
├── models/
└── README.md

Keeping your data, scripts, and generated outputs in one project folder makes it much easier for Codex or Claude Code to understand your workflow.

Step 2: Create a Python Virtual Environment

Open the VS Code integrated terminal:

Use Terminal -> New Terminal
Or press Ctrl + backtick

Then create a virtual environment inside the project: 🖥 Terminal

python3 -m venv .venv
source .venv/bin/activate

If VS Code prompts you to select a Python interpreter, choose the one from your local .venv directory. You can also manually choose it:

Open the Command Palette (Cmd + Shift + P)
Search for Python: Select Interpreter
Select the interpreter inside .venv

Step 3: Install the KumoRFM SDK

Install the SDK inside the active virtual environment: 🖥 Terminal

pip install kumoai

Verify the installation: 🖥 Terminal

python -c "import kumoai.rfm as rfm; print('Kumo SDK loaded successfully')"

Step 4: Authenticate the KumoRFM SDK

You need an API key before you can make calls to KumoRFM. There are two common approaches:

use interactive authentication
set KUMO_API_KEY manually

Interactive authentication Create a small setup script such as scripts/setup_kumo.py:

import os
import kumoai.rfm as rfm

if not os.environ.get("KUMO_API_KEY"):
    rfm.authenticate()

rfm.init(api_key=os.environ.get("KUMO_API_KEY"))
print("Kumo SDK initialized successfully")

Run it from the integrated terminal: 🖥 Terminal

python scripts/setup_kumo.py

Manual authentication

import os
import kumoai.rfm as rfm

os.environ["KUMO_API_KEY"] = "YOUR_API_KEY_HERE"
rfm.init(api_key=os.environ.get("KUMO_API_KEY"))

Step 5: Optional Dependencies

Install Graphviz if you want to visualize graphs from a script-based workflow. Install the system dependency: 🖥 Terminal

brew install graphviz

Install the Python package: 🖥 Terminal

pip install graphviz

Test that a simple graph renders:

from graphviz import Digraph

g = Digraph()
g.edge("Users", "Orders")
g

Part 2: Using the VS Code CLI Workflow

The VS Code integrated terminal is one of the easiest ways to work with the KumoRFM SDK in an editor-based workflow. Instead of switching between separate apps, you can keep all of the following in one place:

your Python scripts
your terminal commands
your coding agent chat panel
project files such as README.md, config files, and sample queries

Common tasks you can run directly from the integrated terminal include: 🖥 Terminal

source .venv/bin/activate
python scripts/setup_kumo.py
python scripts/run_prediction.py

This workflow works especially well when:

you want reproducible scripts instead of interactive notebook cells
you want Codex or Claude Code to edit source files directly
you want to commit your work to Git as normal Python code
you want to run the same setup locally, in CI, or on another machine

A good pattern is to keep small entry-point scripts in scripts/ and ask the coding agent to edit those files rather than generating one-off terminal commands each time.

Part 3: Add a Coding Agent (Optional)

Choose one of the following to add an AI coding agent to your workflow:

Flow A - OpenAI Codex
Flow B - Anthropic Claude Code

Step 1: Install the VS Code Extension

Open VS Code Extensions (Cmd + Shift + X)
Search for “Codex - OpenAI’s coding agent”
Install the extension

Authenticate:

Sign in with ChatGPT (recommended)
Or configure ~/.codex/config.toml

Open Codex using:

the ChatGPT icon in the top-right corner added by the extension

Reload VS Code if the extension does not appear immediately.

Step 2: Install the Kumo Coding AgentThe Kumo Coding Agent has two parts:

Context (knowledge base): documentation, PQL rules, workflow guides, and connector references that teach the agent how to use the Kumo platform
Skills (slash commands): actions like /kumo-issue and /kumo-pr for reporting bugs and contributing fixes

Install the context🖥 Terminal

cd your-project
git clone https://github.com/kumo-ai/kumo-coding-agent.git kumo-coding-agent

This action adds a directory named kumo-coding-agent to your project. It contains the Kumo Coding Agent’s knowledge base. Confirm that this directory appears in your project.Codex reads AGENTS.md automatically. No extra configuration is required.Install the skills (optional) inside a Codex session:🤖 Codex

$skill-installer install https://github.com/kumo-ai/kumo-coding-agent

Step 3: Use Codex from VS CodeCodex works especially well in VS Code when you ask it to edit files in your project instead of isolated notebook cells.Good requests include:🤖 Codex

Create a scripts/run_prediction.py file that loads the RelBench F1 dataset,
builds a Kumo graph, and predicts whether each driver will finish in the
top 3 in the next race.

You can also use Codex to:

add or update Python scripts
create requirements.txt or README.md files
draft PQL queries
help debug terminal errors from your local environment

The integrated terminal is still where you run Python scripts and other shell commands. Codex helps generate and edit the files, but you remain in control of executing the workflow locally.

Step 4 (Optional): UpgradeUpgrade the KumoRFM SDK🖥 Terminal

pip install --upgrade kumoai

Upgrade the Kumo Coding Agent🖥 Terminal

cd kumo-coding-agent && git pull

Step 1: Install the VS Code Extension

Open VS Code Extensions (Cmd + Shift + X)
Search for “Claude Code” (by Anthropic)
Install the extension

Authenticate:

Open Claude Code using either:

the Anthropic spark icon in the sidebar
the Anthropic spark icon in the top-right corner
Cmd + Shift + P -> Claude Code: Open

Reload VS Code if the extension does not appear immediately.

Step 2: Install the Kumo Coding AgentThe Kumo Coding Agent has two parts:

Context (knowledge base): documentation, PQL rules, workflow guides, and connector references that teach the agent how to use the Kumo platform
Skills (slash commands): optional helpers installed with npx skills add

Install the context🖥 Terminal

cd your-project
git clone https://github.com/kumo-ai/kumo-coding-agent.git kumo-coding-agent
echo 'Also read kumo-coding-agent/CLAUDE.md for Kumo agent capabilities.' >> CLAUDE.md

This action adds a directory named kumo-coding-agent to your project. It contains the Kumo Coding Agent’s knowledge base. Confirm that this directory appears in your project.The second line tells Claude Code to read the agent’s knowledge base when it starts.Install the skills (optional)To run the following command, you need Node.js installed on your system:🖥 Terminal

npx skills add kumo-ai/kumo-coding-agent --agent claude-code

Step 3: Use Claude Code from VS CodeClaude Code works best in VS Code when you ask it to reason over the full project, including scripts, prompts, config files, and output files.Good requests include:✨ Claude

Add a scripts/run_prediction.py example that authenticates with Kumo,
loads a local dataset, defines a graph, and runs a simple predictive query.

You can also use Claude Code to:

draft end-to-end example scripts
explain KumoRFM errors from terminal output
propose cleaner project structure
help turn an ad hoc experiment into reusable Python code

Claude Code can update your project files, but you should still run the scripts yourself from the integrated terminal so you can confirm the local environment and outputs are correct.

Step 4 (Optional): UpgradeUpgrade the KumoRFM SDK🖥 Terminal

pip install --upgrade kumoai

Upgrade the Kumo Coding Agent🖥 Terminal

cd kumo-coding-agent && git pull

Step 4: Verify the Setup

Create a minimal script such as scripts/run_prediction.py:

import os
import kumoai.rfm as rfm

if not os.environ.get("KUMO_API_KEY"):
    raise RuntimeError("KUMO_API_KEY is not set")

rfm.init(api_key=os.environ.get("KUMO_API_KEY"))
print("Kumo SDK initialized successfully")

Run it: 🖥 Terminal

python scripts/run_prediction.py

Run Your First Example

Once the SDK and coding agent are set up, ask the agent to help you scaffold a simple project-based workflow. Examples:

create a LocalGraph from local data files
draft a first predictive query in Python
create a reusable script instead of a one-off notebook
add logging, comments, and lightweight validation

If you prefer to start from a guided SDK example first, continue with:

Troubleshooting

VS Code cannot find the interpreter

Re-run Python: Select Interpreter
Make sure .venv exists in the project folder
Restart VS Code after creating the environment

The SDK will not import

🖥 Terminal

source .venv/bin/activate
pip install kumoai

Then rerun your script.

The API key is missing

Export the API key before running a script: 🖥 Terminal

export KUMO_API_KEY="YOUR_API_KEY_HERE"

Then rerun:

python scripts/run_prediction.py

The coding agent edited files, but the workflow still fails

Re-run the script from the integrated terminal
Check that the active terminal is using .venv
Review the modified files before rerunning commands
Ask the agent to fix the error using the exact terminal output

Next Steps

Setup for SDK fundamentals
Make Predictions for PQL query reference
Kumo Coding Agent for agent source and skills

Get Started

KumoRFM SDK

KumoRFM Fine-Tuning SDK

KumoRFM Fine-Tuning Platform

VS Code + Kumo Agent

Overview

Prerequisites

Part 1: Setup for VS Code + KumoRFM SDK

Step 1: Create or Open a Project Folder

Step 2: Create a Python Virtual Environment

Step 3: Install the KumoRFM SDK

Step 4: Authenticate the KumoRFM SDK

Step 5: Optional Dependencies

Part 2: Using the VS Code CLI Workflow

Part 3: Add a Coding Agent (Optional)

Step 4: Verify the Setup

Run Your First Example

Troubleshooting

VS Code cannot find the interpreter

The SDK will not import

The API key is missing

The coding agent edited files, but the workflow still fails

Next Steps

​Overview

​Prerequisites

​Part 1: Setup for VS Code + KumoRFM SDK

​Step 1: Create or Open a Project Folder

​Step 2: Create a Python Virtual Environment

​Step 3: Install the KumoRFM SDK

​Step 4: Authenticate the KumoRFM SDK

​Step 5: Optional Dependencies

​Part 2: Using the VS Code CLI Workflow

​Part 3: Add a Coding Agent (Optional)

​Step 4: Verify the Setup

​Run Your First Example

​Troubleshooting

​VS Code cannot find the interpreter

​The SDK will not import

​The API key is missing

​The coding agent edited files, but the workflow still fails

​Next Steps

Overview

Prerequisites

Part 1: Setup for VS Code + KumoRFM SDK

Step 1: Create or Open a Project Folder

Step 2: Create a Python Virtual Environment

Step 3: Install the KumoRFM SDK

Step 4: Authenticate the KumoRFM SDK

Step 5: Optional Dependencies

Part 2: Using the VS Code CLI Workflow

Part 3: Add a Coding Agent (Optional)

Step 4: Verify the Setup

Run Your First Example

Troubleshooting

VS Code cannot find the interpreter

The SDK will not import

The API key is missing

The coding agent edited files, but the workflow still fails

Next Steps