Sources and loaders

A string source recursively scans a Markdown directory. An object source selects a format explicitly:

source: { type: "json", directory: "./content/authors" }

File formats

Markdown frontmatter becomes metadata, with its body captured separately. JSON, YAML, and TOML sources contain one record per file. Use file() for a file containing multiple entries:

import { file } from "contentgraph/loaders";

const authors = defineCollection({
  loader: file("./content/authors.json"),
  schema: schema.object({ name: schema.string() }),
});

Here, defineCollection and z use the imports from collections and schemas.

IDs and custom sources

Built-in paths resolve from the application root. Default IDs derive from relative filenames, with a frontmatter slug override and trailing /index removal.

During development, a declared file or glob directory can be created after watching starts. Initially missing paths retain a metadata polling fallback to recover if their parent's creation notification is lost; ContentGraph then subscribes to the discovered content normally. Files inside a glob collection are not individually polled.

ContentGraph also accepts native Astro-compatible content loaders. Custom loaders implement load(context) and write entries into context.store. See the collection and loader reference.