ucatlas

Not affiliated with or endorsed by Databricks.

Draws a Databricks Unity Catalog, or a snapshot of one, as a static site: an ER diagram and a lineage graph for each schema, and a list of its objects with their columns, keys, comments and column lineage. It reads the catalog with ucmodel and the lineage API, over REST, so it needs no SQL warehouse. It also serves a local UI over the commands.

uv run ucatlas --profile my-profile draw --catalog my_prod --schema shop -o site
uv run ucatlas draw --snapshot snapshots/my_prod-20261101T120000Z-3f2a9c01b7de.json -o site
python3 -m http.server --bind 127.0.0.1 --directory site 8000   # then open localhost:8000

--snapshot draws a file that ucmodel snapshot wrote, with no workspace and no credentials. Without --schema it draws every schema but information_schema, or every schema in the snapshot. --no-lineage skips column lineage, which takes one API call per column.

The local UI

uv run ucatlas serve [--snapshots snapshots] [--project .] [--port 8765]

A UI for one person, on 127.0.0.1 only. It prints an address with a token, keeps the token in a cookie, and refuses requests without it or for another host name. It lists the snapshots and compares, draws or plans against one. It takes a new snapshot with a profile you choose; it never picks one. With ucstate installed and a project, it plans the spec against a snapshot and shows the commands that would settle each blocked change, as buttons.

Each button runs one command as a subprocess and shows its command line and output. The commands are ucmodel snapshot and diff, ucatlas draw, and ucstate plan --against, rename-table, rename-column, allow and transition --against. Only ucmodel snapshot reaches a workspace, and it only reads. The ucstate commands write one entry or one script into the project, for review in git. Drawings and previews are kept in .ucatlas/, by snapshot hash.

It needs Graphviz's dot on the PATH: brew install graphviz, or apt-get install graphviz. The plan is the repo's docs/diagrams.md.