quickdbd-cli (qdbd)

Compile QuickDBD-syntax schema files into interactive HTML ER diagrams, with no table limit, from the command line. Built to stay readable at 200+ tables.

Customer
-
CustomerId PK int
Name string

Order
-
OrderId PK int
CustomerId int FK >- Customer.CustomerId

pipx install quickdbd-cli        # or: uvx --from quickdbd-cli qdbd ...
qdbd build schema.txt -o schema.html

Docs: https://markphamm.github.io/quickdbd-cli/ · Live demo (200 tables): https://markphamm.github.io/quickdbd-cli/demo/

Not affiliated with QuickDBD. It reads the same text syntax, so schemas move freely between the two.

Features

• Same syntax as QuickDBD: tables, aliases, quoted names, all 9 relationship operators, rel lines, PK/FK/NULL/UNIQUE/INDEX/IDENTITY, default=, doc comments, [settings] and [layout] sections. Error messages are QuickDBD's own.
• Interactive page:
  • explorer sidebar: tables grouped by file or dbt folder, with search (/, Enter jumps to the best match)
  • table detail panel (dbt-docs style, for relationships): description, columns with type/keys/docs, and clickable References / Referenced by lists
  • minimap
  • click a table to focus on it and its neighbours
  • collapse tables: full / keys only / names only
  • hover a column to highlight its relationship
  • hover a dotted-underlined table or column to read its # doc comments
  • PNG/SVG export buttons, and draw.io: an editable diagrams.net file using draw.io's ER table shapes and crow's-foot arrows
  • Path finder: in a table's detail panel, type another table to see the shortest join path, highlighted on the canvas, with a ready-to-paste SQL JOIN (also qdbd path erd/ --from payments --to customers)
  • Regions: a shaded, labelled area per file or dbt folder (toolbar toggle). --by-group on build / export / layout arranges tables so each folder sits together
  • shareable links: the URL keeps the filter, selected table, collapse mode and zoom; Link copies it
• Live editing: qdbd serve reloads when you save, has a browser editor pane with autosave, and remembers where you drag tables.
• Many files, one diagram: point it at a folder and foreign keys can cross files.
• Pip-only install: no Graphviz, Node or browser needed. Layout uses ELK, running in an embedded V8.
• CI-friendly: builds are deterministic, and errors are file:line: error: … with exit code 1.

From a dbt project

dbt docs generate                                   # optional: adds real column types (catalog.json)
qdbd import dbt target/manifest.json -o erd/        # one .txt per folder (staging, marts, sources, ...)
qdbd build erd/ -o erd.html

• Tables: models, seeds and snapshots of your project, plus sources.
• Relationships: relationships tests (including dbt_utils relationships_where) and foreign_key constraints.
• Primary keys: primary_key constraints, or else a column with both unique and not_null tests.
• Docs: model and column descriptions become doc comments, shown as tooltips.
• Options:
  • --select 'marts.*' narrows by name, fqn or path (repeatable)
  • --package all includes installed packages
  • --no-sources drops sources
  • -o schema.txt writes a single file

Re-running the import overwrites the generated files. Table positions survive because they live in the layout file, keyed by table name.

Selecting tables

Like dbt's --select, but it follows relationships instead of lineage. It works in the page's filter box and as -s/--select / --exclude on build and export.

qdbd build erd/ -s 'group:marts' --exclude '*_tmp' -o marts.html
qdbd export erd/ -s '+orders+1' -o orders.png

Schema diffs

qdbd diff old.txt new.txt                         # two versions
qdbd diff --base origin/main erd/                 # git ref vs working tree
qdbd diff --base origin/main erd/ --format markdown -o diff.md   # for a PR comment
qdbd build erd/ --diff-base origin/main -o erd.html              # colour what changed

diff reports tables added and removed, column changes (added, removed, type and flag changes) and relationships added and removed. --format json is available, and --exit-code exits 1 when anything changed.

With --diff-base, the page:

• marks new tables green and changed tables amber, with a bar on each changed column;
• shows a summary banner, and its Show changes link filters to the changes plus their neighbours;
• lists each table's changes in the detail panel.

The selector diff:modified (or diff:added / diff:changed) works like dbt's state:modified, e.g. -s 'diff:modified+1'.

Commands

qdbd check  PATH... [--strict]                 # validate; "No errors found. Good job!"
qdbd build  PATH... -o diagram.html            # shareable interactive HTML
qdbd serve  PATH... [--port 8765]              # live editor + drag-to-arrange
qdbd export PATH... -o out.png [--focus Customer --depth 1]   # PNG, SVG or .drawio
qdbd layout PATH... [--reset]                  # (re)write the layout file
qdbd import dbt target/manifest.json -o erd/   # generate schema files from a dbt project
qdbd path PATH... --from A --to B              # shortest join path + SQL JOIN snippet

PATH is a .txt file or a folder. Every *.txt under a folder is read recursively.

Table positions

The first build auto-arranges the tables. In qdbd serve, drag tables where you want them. Positions are saved to a sidecar file you can commit:

• schema.txt → schema.layout.json
• a folder → <folder>/qdbd.layout.json
• override either with --layout FILE

Once a layout exists, a new table is placed next to the tables it's related to, and nothing else moves. Run qdbd layout PATH --reset to re-arrange everything from scratch. A pasted schema's QuickDBD [layout] section is used for any table the sidecar doesn't cover.

qdbd build never writes the sidecar, so CI builds leave the working tree clean.

Differences from QuickDBD

qdbd accepts a few things QuickDBD rejects, and prints a warning for each one:

• tab characters
• decimal(10, 2) (a space inside a type)
• an explicit NOT NULL
• indented table names
• [layout] entries for unknown tables

qdbd check --strict turns these warnings into errors, so a file that passes is guaranteed to paste into QuickDBD.

default= values keep their case, where QuickDBD lowercases them.

CI

GitHub Action: checks the schema and builds the diagram. On pull requests it also posts one schema-diff comment, updated on every push, and highlights the changes in the diagram artifact.

name: ERD
on: [pull_request, push]
jobs:
  erd:
    runs-on: ubuntu-latest
    permissions: { contents: read, pull-requests: write }
    steps:
      - uses: actions/checkout@v4
      - uses: MarkPhamm/quickdbd-cli@v0.1.5
        with:
          path: erd/                 # schema file(s) or folder(s)
          output: site/index.html

Outputs are html, changed and diff-file.

Publishing: to publish the diagram on GitHub Pages, add actions/upload-pages-artifact + actions/deploy-pages after it. See .github/workflows/erd.yml, which publishes this repo's demo.

pre-commit:

repos:
  - repo: https://github.com/MarkPhamm/quickdbd-cli
    rev: v0.1.5
    hooks:
      - id: qdbd-check
        files: ^erd/.*\.txt$   # point at your schema files

GitLab CI:

erd:
  image: python:3.12
  script:
    - pip install quickdbd-cli
    - qdbd check erd/
    - qdbd build erd/ -o public/index.html
  artifacts:
    paths: [public/index.html]

Development

uv venv && uv pip install -e ".[dev]"
.venv/bin/pytest
python scripts/gen_schema.py        # regenerate examples/synthetic (200 tables)
python scripts/gen_font_metrics.py  # regenerate font width table

Design notes: docs/superpowers/specs/2026-10-17-qdbd-design.md.

Releasing

Put PyPI credentials in .env (gitignored):

TWINE_USERNAME=__token__
TWINE_PASSWORD=pypi-...

Bump version in pyproject.toml and src/quickdbd_cli/__init__.py, commit, then run PATH="$PWD/.venv/bin:$PATH" scripts/release.sh. It builds, uploads to PyPI, tags vX.Y.Z and creates the GitHub release.

License

Apache-2.0. Bundled third-party files are listed in NOTICE.