Marrow

Rheo projects treat .marrow.typ as a special file that allows you to programmatically generate extra files beyond the one-page-per-vertebra default. In contrast to vertebrae, which are treated as normal Typst compiled to PDF or HTML, marrow is Typst run at the bundle root, meaning that you can write Typst in it to generate new documents and assets. You can use marrow to programmatically derive artifacts such as feeds, sitemaps, search indexes, and generated pages.

Crafting marrow

Create a file named .marrow.typ at the top of your content directory, next to your ordinary pages. Inside this file (and only there), document() and asset() are in scope:

#document("extra/hello.html", format: "html", title: [Extra])[Hello from the bundle root.]
#asset("extra/hello.txt", "root-level asset")

Compiling this produces extra/hello.html and extra/hello.txt in your build output, sitting alongside whatever your ordinary vertebrae produce. For more information on what is valid Typst in marrow, see the Typst bundle documentation.

If you’d rather call the file something else, set the top-level marrow key in rheo.toml, resolved against content_dir in rheo.toml:

marrow = "bundle-root.typ"

Note that marrow will never produce a .marrow.html of its own, and it never appears in the spine, the sidebar navigation, or a template’s prev/next pager.

Marrow runs once per per-page output format on every compile and every dev-server rebuild. The combined PDF target skips it entirely, because document() and asset() both hard-error there. Typst reserves them for the bundle target, and a combined PDF has no bundle to speak of.

A page marrow mints is still a first-class Rheo page in every other sense. It receives the same head-asset injection as an ordinary vertebra, and any root-level #set/#show rule from your template still applies inside it.

Reading across the spine

A Rheo bundle compiles in a single Typst pass, and Typst’s own bundle mechanism shares one introspector across every file in it. That means marrow can read state(...) and query labels registered by ordinary vertebrae elsewhere in the project. Marrow thus lets a package or project turn data scattered across many pages into pages of its own, without a stub .typ file per entry:

// across various vertebrae in the spine...
#state("notes", ()).update(old => old + (my-note,))

// in .marrow.typ
#context {
  for n in state("notes", ()).final() {
    document("notes/" + n.name + ".html", format: "html", title: [Note])[#n.body]
  }
}

Typst has no facility for enumerating files on disk, so marrow can never discover content by scanning your project directory itself—only by reading state, labels, or sys.inputs.rheo-context, all of which have to have been populated by something else in the bundle first.