Writing SQL

In trend, a dataset is a saved SQL query — the unit of reuse across dashboards. Every chart binds to a dataset.

The editor

The dataset editor gives you schema-aware autocomplete. Type a table or column name and you’ll get completions from the schema cache for the connection you picked.

SELECT
  date_trunc('week', created_at) AS week,
  count(*) AS signups
FROM users
GROUP BY 1
ORDER BY 1

Click Preview to run against the connection. The result panel shows columns and rows; we cap previews at 10,000 rows by default.

Workspace vs. local datasets

• Workspace datasets live in the Datasets section. Any dashboard in the workspace can use them. Best for queries you’ll reuse — "Active users", "Daily revenue", "Top regions".
• Local datasets live inside a single dashboard. Best for one-off queries that don’t deserve to clutter the workspace library.

You can promote a local dataset to workspace-scope from its menu.

What makes a good dataset

• One unit of meaning per dataset. "Active users by week" is one. Don’t cram active users AND signups AND churn into one query just because they share a FROM clause.
• Templated parameters where appropriate. See Templates.
• Avoid heavy aggregations the chart could do. Return the raw grain when feasible — chart-level aggregations let one dataset feed multiple chart angles.

What to read next

• Templating with variables — the {{var}} and [[clause]] syntax.