Completely redesigned for v0.12.0 — master-detail layout, three-component pipeline editor, and a universal SQL execution engine
The recipe editor uses a two-panel layout: a scrollable recipe list on the left and the editor on the right. The old single-step interface has been removed. Recipes are accessible via deep linking at /recipes/[id], making it easy to share direct links to specific recipes.
Each recipe is built from three components displayed as collapsible cards:
| Component | Purpose |
|---|---|
| Dataset / Transform | Select the input dataset and apply column-level transformations — 8 operation types including rename, cast, derive, filter, sort, aggregate, join, and union. Live preview shows results at each step. |
| Algorithm | Choose the analysis method — descriptive statistics, correlation, t-tests, ANOVA, chi-square, linear regression, or custom SQL. Algorithm types are discovered dynamically from the backend API. |
| Output | Select the visualisation type and configure chart options. Output types are also discovered dynamically from the backend API. Supports bar, line, scatter, area, horizontal bar, donut, histogram, distribution, and pie charts. |
The components are linked with bidirectional constraint propagation — changing the dataset type updates which algorithms are available, and selecting an algorithm constrains which outputs make sense.
Every recipe has a unique URL at /recipes/[id]. This means you can bookmark specific recipes, share links with colleagues, and navigate directly to a recipe without going through a list page. The recipe ID is stable and persists across edits.
Recipes execute through the SQL Execution Engine, a universal ops compiler that translates recipe stages into SQL for five database backends: DuckDB, PostgreSQL, MySQL, MSSQL, and MS Access. When a backend lacks native support for an operation, the engine falls back to pandas automatically.
Execution results include a Kill button that lets you cancel a running recipe. Results are shown on the execution results page with statistics and visualisation output.
Recipes can be shared with your user group via a share toggle. Shared recipes display a "Shared by" badge. Non-owners view shared recipes in read-only mode. Any recipe can be forked (cloned) to create your own copy — useful for building on a colleague's work without affecting the original.