@vllnt/next-llms

Make your Next.js site legible to AI agents — generate a spec-compliant llms.txt manifest and serve any page as raw markdown.

// app/llms.txt/route.ts
import { generateLlmsText } from "@vllnt/next-llms";

export function GET() {
  return new Response(
    generateLlmsText({
      title: "Acme",
      summary: "Acme is a widget API.",
      sections: [
        {
          title: "Docs",
          links: [{ title: "Quickstart", url: "/docs/quickstart.md" }],
        },
      ],
    }),
    { headers: { "Content-Type": "text/plain; charset=utf-8" } },
  );
}

Emit the llms.txt index and expose any page's markdown at /path.md so agents skip HTML parsing. Zero runtime dependencies — pure Web-standard APIs that run in any Next.js runtime (and Remix, Hono, Deno, …).

Features

• Spec-compliant llms.txt — generateLlmsText emits the llmstxt.org shape: an # title, a > blockquote, ## Section link lists, and a droppable ## Optional section.
• Raw-markdown routes — createMarkdownRoute serves any page as text/markdown, stripping a trailing .md and resolving the slug through your own content source.
• Zero dependencies — pure TypeScript over Web Request / Response; nothing added to your bundle.
• Runtime-agnostic — the route handler imports nothing from next; it runs on the Edge, Node, or any Web-standard runtime.
• Typed end to end — concrete config types guide correct usage; no any.
• Cache-friendly — sensible Cache-Control defaults, overridable per route.

Installation

pnpm add @vllnt/next-llms@canary

Node.js >=18; no peer dependencies. This README follows current main and the canary channel (0.1.0-canary.3cfe2be at this audit). npm latest also points to a prerelease, but an older one (0.1.0-canary.bca5ab7), so an untagged install does not select current main. There is no stable release on latest yet. Pin the resolved prerelease and test upgrades. Current main removed the older generateLlmsFullText API; use the two exports below.

Usage

// app/[...slug]/route.ts — serve any page as raw markdown
import { createMarkdownRoute } from "@vllnt/next-llms";

export const GET = createMarkdownRoute(async (slug) => {
  const doc = await getDoc(slug.join("/")); // your content source
  return doc?.markdown ?? null; // null → 404
});

/docs/intro.md resolves the slug ["docs", "intro"] and returns the markdown with Content-Type: text/markdown, or 404 when the resolver returns null.

API Reference

Export	Kind	Result
generateLlmsText(config)	function	string — the llms.txt index
createMarkdownRoute(resolver, options?)	factory	(request: Request) => Promise<Response>

Full reference: docs/API.md. Also see the LLM index, changelog, and issue tracker.

Security

• No secrets touched — the package generates text and serves whatever your resolver returns.
• Your resolver is the boundary — validate and authorize the slug before returning content.
• Successful responses default to public caching (one hour in browsers, one day in shared caches). For private or personalized content, explicitly pass { cacheControl: "private, no-store" }; authorization alone does not disable caching.

See docs/API.md and the security reporting policy.

Testing

pnpm test           # single run
pnpm test:coverage  # enforced 100% on covered files

Contributing

See CONTRIBUTING.md.

Author

Built by bntvllnt · bntvllnt.com · X @bntvllnt

Part of the @vllnt open-source fleet — vllnt.com

If this is useful, sponsor the work.

License

MIT — see LICENSE.