Generate artifacts without a docs tree

Use this guide when a site has no .mdx docs tree but should still be readable by agents: a CMS-backed blog, a data-driven site (benchmarks, changelogs, directories), a marketing site, or a microfrontend that owns one slice of a larger origin.

The docs pipeline reads pages from .md/.mdx files on disk. generateAgentArtifacts() skips that step: you hand it a page list built from any source — a CMS API, a database, a route manifest — and it emits the same artifact set the docs CLI produces.

Generate the artifact set

Call it from a build script, after your content is available:

import { generateAgentArtifacts } from "leadtype/llm";

await generateAgentArtifacts({
  outDir: "public",
  baseUrl: "https://cookiebench.com",
  product: {
    name: "CookieBench",
    tagline: "Benchmarks for consent banner performance.",
  },
  organization: { name: "Consent.io", url: "https://consent.io" },
  agents: { robots: { policy: "block-training" } },
  groups: [{ slug: "benchmarks", title: "Benchmarks" }],
  pages: benchmarks.map((benchmark) => ({
    urlPath: `/benchmarks/${benchmark.slug}`,
    title: benchmark.name,
    description: benchmark.summary,
    content: renderMarkdown(benchmark),
    lastModified: benchmark.updatedAt,
    groups: ["benchmarks"],
  })),
});

This writes into outDir:

• /llms.txt (plus the /.well-known/llms.txt discovery copy) — product summary, your authored llms.sections, and a grouped page map linking every page's .md mirror.
• One markdown mirror per page at ${urlPath}.md (the root page / becomes /index.md), with the frontmatter agents need for attribution and staleness checks: title, description, canonical_url, last_updated. Mirrors pass the agent-readability frontmatter checks when served statically — no runtime enrichment required.
• /sitemap.xml (with lastmod), /sitemap.md, and /robots.txt with the Content-Signal policy you chose.
• /agent-readability.json — the same manifest shape the docs pipeline emits, written at the site root instead of under /docs/. Point the runtime helpers from Serve agent responses at it to add Accept: text/markdown content negotiation and per-page JSON-LD.

Page inputs

Each page needs a urlPath and markdown content. Everything else falls back:

• title — explicit value, then frontmatter title: inside content, then the URL slug.
• description — explicit value, then frontmatter description:.
• lastModified — explicit string | Date, then frontmatter last_updated: / lastModified:, then the generation time. Pass real content dates when you have them; agents use last_updated to judge staleness, and sitemap.xml reuses it for lastmod.
• groups / order — match the groups config exactly like frontmatter group: and order: in the docs pipeline. Ungrouped pages render under ## Other (or ## Pages when no groups are configured).

Frontmatter already present in content is parsed and stripped — explicit fields win, and the mirror is written with a single normalized frontmatter block.

Duplicate urlPaths, paths without a leading /, and paths containing ? or # throw before anything is written.

Microfrontends: merge fragments at the host

Root crawler files (robots.txt, sitemap.xml, sitemap.md) are origin-level — only one app on the domain should serve them. When a blog or docs app deploys under a host app's domain, generate its fragment with:

await generateAgentArtifacts({
  // ...
  emitRootCrawlerFiles: false,
});

The fragment still gets its llms.txt, mirrors, and manifest; the host app owns the root files and merges fragment manifests into its own sitemap.xml and llms.txt. The docs pipeline's artifacts are scoped under /docs/ for the same reason.

When to use the docs pipeline instead

If your content already lives in .md/.mdx files, use the static artifact CLI — it adds MDX flattening, i18n, transformers, search artifacts, and skills on top of this same output. generateAgentArtifacts() is the escape hatch for content that never touches the filesystem.