semql-erd

ER-diagram generator for semql catalogs. Walks the cubes and joins in a Catalog and emits a Graphviz DOT source (and convenience PNG/SVG when the graphviz Python bindings

• system dot binary are available).

Useful when:

• The catalog is past 10 cubes and reading the YAML/Python isn't enough to see the join shape at a glance.
• A PR touches a Join and the reviewer wants a visual diff of the before / after graph.
• Onboarding docs need a stable picture of what's in scope.

Install

pip install semql-erd            # DOT source only — no system deps
pip install "semql-erd[image]"   # + graphviz Python bindings
                                 #   (also needs the `dot` binary)

Quick start — DOT source

render_dot(catalog) is dependency-free: it produces a DOT-language string you can paste into any Graphviz renderer (Edotor is a quick web one).

from semql import Dialect, Catalog, Cube, Dimension, Join, Measure
from semql_erd import render_dot

orders = Cube(
    name="orders",
    dialect=Dialect.POSTGRES,
    table="orders",
    alias="o",
    measures=[Measure(name="revenue", sql="{o}.amount", agg="sum", unit="currency")],
    dimensions=[Dimension(name="region", sql="{o}.region", type="string")],
    joins=[Join(to="customers", relationship="many_to_one", on="{o}.cid = {c}.id")],
)
customers = Cube(
    name="customers",
    dialect=Dialect.POSTGRES,
    table="customers",
    alias="c",
    dimensions=[Dimension(name="name", sql="{c}.name", type="string")],
)

print(render_dot(Catalog([orders, customers])))

Quick start — PNG/SVG

from semql_erd import render_image

# Requires `pip install "semql-erd[image]"` AND the `dot` binary on PATH.
render_image(catalog, "catalog.png")  # PNG by default
render_image(catalog, "catalog.svg", format="svg")

Conventions

• Nodes are cubes. The label is a Graphviz record showing the cube name (+ display_name suffix if set), the backend, and three field sections (measures, dimensions, time-dimensions).
• Edges are Joins. Arrowhead shape encodes the relationship:
  • many_to_one → crow on the from-side, tee on the to-side
  • one_to_many → mirror of the above
  • one_to_one → tee on both sides
• Filtering mirrors the planner prompt: by default only cubes with expose_in_prompt=True (and non-META cubes) appear. Pass only_exposed=False for a full graph.
• Layout defaults to rankdir="LR" (left-to-right). Pass rankdir="TB" for top-to-bottom.

CLI

python -m semql_erd path.to.module:catalog          # prints DOT to stdout
python -m semql_erd path.to.module:catalog out.svg  # writes a rendered image

Status

Early development. The DOT format is stable; record-section ordering and node ID naming may evolve.