shelves-bi 0.3.0

AI-native declarative visual analytics — YAML DSL to Vega-Lite translator

Declarative visual analytics for semantic models.
Write charts and dashboards in YAML, render to Vega-Lite.

Shelves charts read from a semantic model — measures, dimensions, formats, and aggregations defined once and reused everywhere, so you never copy-paste field logic across dashboards. You can back that model two ways:

• Start with a flat file. Point a model at a local CSV, Parquet, or JSON file and Shelves queries it directly with DuckDB. shelves-import can even generate the model for you. Zero infrastructure — go from a raw file to a chart in seconds.
• Grow into a semantic layer. Point the same model at a Cube.dev instance when you want shared governed definitions across a team. Only the model's source block changes; your charts stay untouched.

The two paths use the identical model and chart schema, so a project can graduate from file to Cube without rewriting a single chart.

Install

pip install shelves-bi

For the flat-file path, install the optional DuckDB extra:

pip install 'shelves-bi[duckdb]'

Requires Python 3.11+.

Quick start (flat file)

Generate a model from a CSV — string columns become dimensions, numeric columns become measures, dates become temporal dimensions:

shelves-import sales.csv          # writes models/sales.yaml

The generated model points at your file:

# models/sales.yaml
model: sales
label: Sales

source:
  type: file
  path: sales.csv

measures:
  revenue:
    column: Revenue
    aggregation: sum
    label: Revenue
    format: "$,.0f"

dimensions:
  category:
    column: Category
    label: Category
  order_date:
    column: Order Date
    type: temporal
    label: Order Date
    defaultGrain: month

Write a chart that references the model by name, then render:

# charts/revenue_by_category.yaml
sheet: "Revenue by Category"
data: sales

cols: category
rows: revenue
marks: bar
color: category
sort:
  field: revenue
  order: descending

shelves-render charts/revenue_by_category.yaml --models-dir models/

Output goes to output/<sheet-name-slug>.html by default. Use --out to override.

Connect to Cube

When you're ready for a governed semantic layer, point the model's source at a Cube.dev instance instead. Set your credentials in a .env file or as environment variables:

CUBE_API_URL=http://localhost:4000
CUBE_API_TOKEN=your-cube-api-token

Only the source block changes — measures, dimensions, labels, and formats are declared the same way as the file model above:

# models/orders.yaml
model: orders
label: Orders

source:
  type: cube
  cube: orders

measures:
  net_sales:
    label: Net Sales
    format: "$,.0f"
    aggregation: sum

dimensions:
  category:
    label: Category
  order_date:
    type: temporal
    label: Order Date
    defaultGrain: month
    format:
      month: "%b %Y"

Your existing charts keep working unchanged.

Project structure

A typical Shelves project looks like this:

my-project/
  models/
    sales.yaml           # semantic model definitions (file or cube source)
  charts/
    revenue_by_category.yaml
    sales_over_time.yaml
  dashboards/
    overview.yaml
  .env                   # CUBE_API_URL / CUBE_API_TOKEN (only for the Cube path)

Dashboards

Dashboards compose multiple charts into a single HTML page with layout, text, and styling:

# dashboards/overview.yaml
dashboard: "Sales Overview"
canvas: { width: 1440, height: 900 }

root:
  orientation: vertical
  contains:
    - text: "Sales Overview"
      preset: title
    - horizontal:
        contains:
          - sheet: "charts/revenue_by_category.yaml"
            width: "60%"
          - sheet: "charts/sales_over_time.yaml"
            width: "40%"

# Dashboard (charts and models resolved from directories)
shelves-render dashboards/overview.yaml --chart-dir charts/ --models-dir models/

# Dev server with live reload
shelves-dev charts/revenue_by_category.yaml --models-dir models/

Python API

from shelves import parse_chart, translate_chart, merge_theme, render_html
from shelves.data.bind import resolve_data

spec   = parse_chart(yaml_string)          # YAML -> ChartSpec
vl     = translate_chart(spec)             # ChartSpec -> Vega-Lite dict
themed = merge_theme(vl)                   # apply default theme
final  = resolve_data(themed, spec)        # query the model's source and bind
html   = render_html(final)                # standalone HTML with vegaEmbed

Each step is independent and composable.

Documentation

• Getting Started — setup, first chart, first dashboard
• DSL Reference — complete field and property reference
• Dashboards — layout DSL, components, and styling

Development

git clone https://github.com/shelveshq/shelves.git
cd shelves
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/pytest

License

Apache 2.0