Skip to content
Saiku Cloud

Schemas

Schemas is the catalog and the front door to authoring. Each schema bundles one or more cubes with their dimensions, measures and joins — and a cube is what people and agents actually query.

The page has two halves. The top offers the ways to start something new; the bottom lists what you already have.

Starting a new schema

New schema opens the visual canvas: pick a warehouse, drop tables on it, define dimensions and measures, save. No AI in the loop — it’s for people who know the shape they want. See Schema designer.

Draft with AI takes a data source and a sentence — “monthly sales by product category and region” — and hands back a complete cube proposal you can edit before saving.

Use a template lives one page over, on Data sources, because a template provisions tables into a specific warehouse.

You can also upload Mondrian XML you already have. Every save, whichever route it came from, becomes a new version under the workspace you chose.

The catalog

The table lists every saved schema with its database, cube count, last-updated time and version. Filter by database narrows it when one warehouse’s cubes are all you care about, and every column header sorts.

Locally-saved drafts appear here too — an unfinished canvas you haven’t committed yet, kept in your browser rather than on our servers. They’re marked as drafts and only you can see them.

Open loads the schema back onto the canvas with its warehouse already selected, so a change is one click from where you left off. Saving from there creates the next version; the previous one is never overwritten.

The row menu

The ⋯ menu on each row is where the less-frequent actions live:

  • View version history — every saved version, newest first, with when it was created, how (uploaded, drafted, template), how many cubes it held, and a route back onto the canvas for any of them. There’s a Who column too, but it stays empty for now — author attribution is waiting on role-based access control.
  • Rename — change the label without touching the definition.
  • Duplicate — fork the schema so you can experiment without risking the version people are querying.
  • Download as Mondrian XML / Mondrian YAML / canvas JSON — three views of the same definition. XML is what the engine loads; YAML is the friendlier form for reviewing in a pull request; canvas JSON round-trips the visual layout.
  • Remove — delete it. There’s no undo, and workbooks pointing at its cubes will stop resolving, so take a download first if there’s any chance you’ll want it back.

What schema XML looks like

You shouldn’t often need to read this — the canvas and the AI draft both write it for you — but a minimal cube is small enough to hold in your head:

<Schema name="sales">
  <Cube name="Sales">
    <Table name="fact_sales"/>
    <Dimension name="Date" foreignKey="date_id">
      <Hierarchy hasAll="true" primaryKey="date_id">
        <Table name="dim_date"/>
        <Level name="Year"    column="year"    type="Numeric"/>
        <Level name="Quarter" column="quarter" type="Numeric"/>
        <Level name="Month"   column="month"   type="Numeric"/>
      </Hierarchy>
    </Dimension>
    <Measure name="Revenue" column="revenue"  aggregator="sum"/>
    <Measure name="Orders"  column="order_id" aggregator="count"/>
  </Cube>
</Schema>

The full reference — every element, every attribute, and the patterns that actually come up — is in Mondrian schemas.