Skip to main content
Oxygen is the open-source AI data analyst built for speed and precision. Written in Rust and declarative by design, Oxygen provides the foundational components needed to transform AI-driven data analysis into reliable, production-ready systems through structured primitives, semantic understanding, and predictable execution. Using Oxygen, users can automate data Q&A and reporting, and accelerate building data artifacts such as semantic topics and data applications. Oxygen integrates natively with your existing data stack - data warehouses, ELT tools, semantic layers, and BI tools. Oxygen also comes with its own data tools for a zero-config experience. Oxygen applies software development lifecycle principles (e.g. build-test-deploy pipeline) to AI-driven data analytics. Oxygen establishes a structured workflow for data agents, involving agent creation, prompt testing, and production deployment. banner Specifically, this process is implemented as follows:
  • Agent Configuration: Define agents using .agent.yml files that specify their instructions. Agents are provided tools to generate SQL, execute semantic queries, and execute procedures. We recommend using a routing agent as the first layer, which can deterministically execute procedures, with a SQL-generation agent as a fallback. This ensures that vetted procedures run with high determinism against their attached inclusions, while the SQL generation fallback provides broad coverage for ad-hoc questions.
  • Procedure Development: Create .procedure.yml files to orchestrate multi-step processes. Use retrieve: include and exclude to control how these procedures are retrieved by agents that have access to them.
  • Testing Framework: Add test cases directly to .agent.yml or .procedure.yml files. Execute tests using the oxy test command.
You can interact with all components — agents, procedures, and data — through the web interface launched by oxy start. Oxygen is also CLI-native, so every operation can be run from the terminal, making it easy to integrate with coding tools like Claude Code and CI/CD pipelines.

​
Who should read what?

The documentation is split into two audiences: End users — data analysts, business users, and developers who use Oxy to query data, build agents, and create workflows. Start here:

Quickstart

Install Oxy and run your first query in minutes

Getting Started with Agents

Create and test your first AI data agent

Core Concepts

Agents, workflows, semantic layer, data apps

Basic Oxy commands

CLI reference for running and testing agents
Operators — DevOps engineers, platform teams, and self-hosters who deploy and configure the Oxy server. Start here:

Deployment Overview

Deploy Oxy on cloud, Docker, or Kubernetes

Deployment Modes

Multi-workspace vs single-workspace mode

Authentication

Magic link, Google OAuth, Okta SSO

GitHub App Setup

Enable GitHub workspace import