​
Overview

In this quickstart guide, we will install oxy and bootstrap a simple setup that uses DuckDB as warehouse and OpenAI as LLM provider. For this walkthrough, you’ll need an OpenAI API key and access to a terminal.

​
Install and configure oxy

Oxy supports both Mac and Windows. To get started with oxy, you will need to work within the terminal. For Mac users, we recommend using iTerm2 as it offers better features, customization options, and more frequent updates as compared to the stock Mac terminal. Other great terminal options include the GPU-accelerated Alacritty or AI-native Warp. For windows, we recommend installing WSL 2 (search forUbuntu in the Microsoft store). Once you have your terminal set up, follow the steps below:
1

Install oxy

Install oxy by running the following within your Mac/Linux/WSL terminal:
bash <(curl --proto '=https' --tlsv1.2 -sSf https://raw.githubusercontent.com/oxy-hq/oxy/refs/heads/main/install_oxy.sh)
Upon successful installation, you should see the message “Oxy has been installed successfully.”You may need to restart your terminal for the command to be recognized.
For transparency, you can review the installation script’s source code.
After the installation is complete, you may verify that it was successful by running:
oxy --version
2

Create a sample repository

In order to get started with a base agent set up, run the following command in an empty directory where you want to store your project:
oxy init
This command will generate the main oxy configuration file (config.yml), a sample agent file, a sample workflow file, some data, and a sample semantic layer file. If you use all the configuration defaults, you should be able to run the agent and workflow against the sample data out-of-the-box.
3

Connect oxy to OpenAI

For the model, we recommend using OpenAI’s gpt-4.1. For alternative models, please refer to our integration documentation, here.Replace OPENAI_API_KEY with your actual OpenAI API key.

​
All Done

Congrats! You’re ready to start using Oxy now. If you haven’t changed the configuration defaults, you should be able to run the following commands to test out oxy against some sample sleep data: 
oxy run sql-generator.agent.yml "On how many nights did I get good sleep in the last year?"  # ask a question to the sample agent
oxy test sql-generator.agent.yml  # runs all defined tests against the sql-generator agent

oxy run report-generator.workflow.yml  # execute the sample workflow
oxy test report-generator.workflow.yml  # run all defined tests against the workflow

oxy serve  # stands up a GUI against your agents and workflows