Skip to content
Saiku Cloud

Sign up & first cube

This page is the long-form walkthrough of the Quickstart, with the why behind each step. If you just want to ship a cube, Quickstart is faster.

Sign up

  1. Visit cloud.saiku.bi/signup.

  2. Sign in with email or Google — we delegate identity through WorkOS, so there’s no Saiku password to invent or forget. (Enterprise customers use SAML through the same door.)

  3. Name your tenant and pick its slug — kebab-case, 2 to 32 characters, and unique across Saiku Cloud. Choose deliberately: the slug is part of your tenant’s identity and can’t be changed later without our help.

  4. Everything else is provisioned for you:

    • The tenant itself, on the slug you chose.
    • A workspace to hold your schemas and workbooks.
    • Your user account, as the tenant owner.
    • An initial API key, so the dashboard can talk to the backend on your behalf. Mint more for agents and integrations on the API keys page whenever you need them.

Connect your first warehouse

Five warehouses you can connect today: PostgreSQL, MySQL / MariaDB, Snowflake, BigQuery and MotherDuck. ClickHouse sits in the picker marked coming soon — the driver is there, we just haven’t opened it. Want a warehouse that isn’t listed? Ask; a new dialect is one JDBC driver plus a small classifier, not a rewrite.

You don’t assemble a JDBC URL yourself. Each dialect has named fields, and Saiku builds the URL from them (with secure defaults like sslmode=require) and previews what it built. An Advanced toggle takes a raw URL for anything unusual.

Host:       db.acme.com
Port:       5432
Database:   analytics
Username:   saiku_reader
Password:   ••••••••

SSL is required automatically — you don’t set it.

Each dialect has its own step-by-step guide under Connecting your warehouse, including the firewall allowlist and the least-privilege read-only role to create.

Credentials are encrypted at rest. See Tenant isolation for how we handle the credentials you give us.

Why a cube?

Saiku speaks MDX over Mondrian XML schemas. A cube is a declarative model that says “here are my facts (a sales table), here are the dimensions I want to slice by (date, region, product), and here are the measures I want to aggregate (revenue, units)”. Once you’ve described that once, any OLAP-style query — drilldowns, pivots, hierarchies — works against it without writing SQL.

Saiku Cloud’s job is to make cube authoring not-a-chore. You either draw the cube on a visual canvas or let AI draft it from a sentence and a look at your warehouse. Either way you review and tweak; you never start from a blank schema file.

Draft → Save → Query

The AI path, end to end:

  1. Pick a data source. Any saved warehouse connection, or a file you’ve uploaded. We profile it cheaply — read the metadata, sample a few rows per column.

  2. Pick a fact table. Likely aggregate tables (agg_*, *_summary, *_rollup) are pushed to the bottom of the list, so the canonical base-grain fact wins rather than getting lost under thirty pre-aggregations.

  3. Describe the cube in plain English. Optional, but it makes a real difference: “Sales facts joined to customer and product dimensions, count of orders and sum of revenue.”

  4. Review the proposal. Measures and dimensions come back as editable cards. Rename them, change an aggregator, drop what you don’t need, correct a join the model guessed wrong.

  5. Preview cube schema. This validates the cards and generates the Mondrian XML — and unlocks three tabs worth using before you commit: Source, Sample data, and Try a query. Run the sample query. A join in the wrong place shows up instantly as inflated totals or a column of nulls.

  6. Save as new schema version. The schema lands in your workspace, bound to the connection.

  7. Analyze. Sidebar → Analyze → Open in Saiku on the cube card. You land in the Saiku UI with the cube selected, ready to query.

Prefer to build it by hand? New schema on the Schemas page opens the visual canvas instead — tables and joins, then dimensions and hierarchies, then measures and calculations, then a validation step. No AI involved, and no LLM budget spent.