Skip to main content
Codex

​
Overview

This guide walks you through setting up the KumoSDK and Kumo Coding Agent with OpenAI Codex in desktop and CLI environments. Codex is an AI coding agent by OpenAI. It can read your project files, write code, run commands, and iterate on errors. You interact with it by typing natural language prompts.

​
Prerequisites

Make sure you have:
  1. Python 3.10 to 3.13 installed
  2. A Kumo account and API key (Create)
  3. An OpenAI account (ChatGPT or API)
  4. Codex installed and authenticated. See the official Codex docs for desktop app and CLI installation.

​
Step 1: Create a Project

Create or open a project directory: 
mkdir my-kumo-project
cd my-kumo-project
Desktop app: Open Codex and select your project folder. CLI: Navigate to your project in the terminal.

​
Step 2: Install KumoSDK

pip install kumoai
Verify: 
python -c "import kumoai; print(f'kumoai {kumoai.__version__} installed')"

​
Step 3: Authenticate with Kumo

Set your API key: 
export KUMO_API_KEY="YOUR_API_KEY_HERE"    # from https://kumorfm.ai/api-keys

​
Step 4: Install Kumo Coding Agent

The Kumo Coding Agent has two parts:
  • Context (knowledge base): Documentation, PQL rules, workflow guides, and data connector references that teach the agent how to use the Kumo platform. Installed by cloning the repository.
  • Skills (slash commands): Actions like /kumo-issue and /kumo-pr for reporting bugs and contributing fixes. Installed via the skill installer.
Install the context: 
git clone https://github.com/kumo-ai/kumo-coding-agent.git kumo-coding-agent
Codex reads AGENTS.md automatically. No extra configuration needed. Install the skills (optional). Open a Codex session (see Step 5) and type: 
$skill-installer install https://github.com/kumo-ai/kumo-coding-agent

​
Step 5: Use the Kumo Coding Agent

Desktop app:
  1. Open the Codex desktop app
  2. Select your project folder
  3. Type a prompt in the chat panel at the bottom:
Load the RelBench F1 dataset and predict whether each driver will finish in the top 3 in the next race
  1. Codex will read the Kumo Coding Agent context, inspect the data, build a graph, write PQL, and run the prediction.
  2. Review the changes Codex proposes and approve them.
CLI: Run a prompt directly from the terminal: 
codex "Load the RelBench F1 dataset and predict whether each driver will finish in the top 3 in the next race"
Codex will generate files, run code, and output results in the terminal.
Codex runs headless in CLI mode. Provide the full prompt upfront. For multi-step workflows, run Codex multiple times or use the desktop app for interactive sessions.

​
Step 6 (Optional): Upgrade

Upgrade KumoSDK: 
pip install --upgrade kumoai
Upgrade the Kumo Coding Agent: 
cd kumo-coding-agent && git pull
To pin to a specific version: 
cd kumo-coding-agent && git checkout v1.0.0

​
Troubleshooting

​
kumoai cannot be imported

pip install kumoai

​
API key is missing

export KUMO_API_KEY="YOUR_API_KEY_HERE"

​
Agent does not respond with Kumo knowledge

  • Verify kumo-coding-agent/ exists in your project directory
  • Verify kumo-coding-agent/AGENTS.md exists
  • Restart Codex or open a new session

​
Codex CLI not found

npm install -g @openai/codex

​
Next Steps