The architecture

Drawn from what's true.

Architecture read out of your repository rather than out of somebody's memory — and re-read whenever the repository changes.

compose, OpenAPI, SQL

The diagram is generated from sources you already maintain, so it describes the system that exists rather than the one that was designed.

Your real config — docker-compose, an OpenAPI spec, a SQL schema, or a live database.
Already maintained — these are files you keep current anyway, so the diagram inherits that.
Not from memory — nobody redraws the architecture by hand, so nobody forgets a service.

You describe what things are and how they relate. Where the boxes go is explicitly not your problem.

You send semantics — lanes, groups, typed nodes, ports and edges — the meaning, not the geometry.
Never coordinates — layout is computed, so a diagram cannot be broken by adding one more node.

Re-generate after a change and the diagram catches up with the code, which is the whole reason it is worth having one.

Refreshable — regenerating is the maintenance story; there is no manual redraw.
Deep links — a node can be linked from the page that explains it, and back again.