Skip to main content
Data Infrastructure is the foundation of Oxygen, getting your data in and shaping it into clean tables that your World Model and agents can query. Four engines do the work:
  • Airway loads: it extracts data from the systems a business already runs and loads it into the warehouse, moving only what changed.
  • Airhouse stores: it is the managed, Postgres-compatible warehouse where that data lands, along with the facts custom apps record, with isolated databases and no long-lived keys.
  • Airform shapes: it turns raw tables into clean, tested ones with version-controlled SQL models.
  • Airconductor runs: it queues each step and runs it in order, so a load, the SQL that shapes what it loaded, and the refresh of the model built on top happen in sequence and finish even if a server restarts partway through.
Data Infrastructure as a left-to-right flow. Sources such as point-of-sale, accounting, databases, files, and public data are extracted by Airway pipelines, which run incrementally from a cursor and load raw tables into a warehouse, either Airhouse or your own. Airform's version-controlled SQL models shape the raw tables into clean, tested tables, which the World Model's semantic model queries for agents and custom apps. Custom apps also write their own facts back into the warehouse. A highlighted band notes that every pipeline, model, and metric is a version-controlled file in one workspace, deployed through Git.

Data moves left to right, from your systems to the World Model, and every step along the way is a version-controlled file in the same workspace.

​
One workspace, from source to model

A typical data stack is stitched together from separate products for moving, storing, transforming, and defining data, each with its own configuration, credentials, and release process. The seams between them are where numbers drift and failures hide. In Oxygen, every step lives in one workspace. Pipelines, models, and metric definitions are files reviewed and deployed together through Git, alongside the agents and automations that use them. A change to how data is loaded, shaped, or measured is a single, reviewable change, and every number an agent reports traces back to definitions you can read.

Airway

Incremental ELT from the systems you already run.

Airhouse

The managed warehouse where your data lands.

Airform

Version-controlled SQL models that land clean, tested tables.

Airconductor

The task engine that runs every load, step, and schedule to completion.