@dualmark/vercel

Vercel Edge Middleware adapter for the Dualmark AEO framework. Wraps any upstream handler and transparently serves markdown to AI bots at the edge — no changes to your existing site.

Install

bun add @dualmark/vercel @dualmark/core

Usage

// proxy.ts
import { NextResponse } from "next/server";
import { createAEOMiddleware } from "@dualmark/vercel";

const middleware = createAEOMiddleware({
  upstream: async () => NextResponse.next(),
  fetchAsset: async (url, init) => fetch(url.toString(), init),
  trailingSlash: "never",
  enableLinkHeader: true,
  hooks: {
    onAIRequest: (info) => console.log(`${info.botName} hit ${info.pathname}`),
    onMiss: (info) => console.warn(`miss: ${info.pathname}`),
  },
});

export default middleware;

export const config = {
  matcher: [
    {
      source: "/((?!_next/|favicon.ico).*)",
      missing: [{ type: "header", key: "next-router-prefetch" }],
    },
  ],
};

How it works on Vercel

Vercel Edge Middleware re-triggers for same-origin fetch() calls. To prevent infinite loops, the adapter adds an x-dualmark-subrequest header to internal fetchAsset calls and short-circuits when it detects a subrequest — returning NextResponse.next() so Vercel serves the static file directly.

• upstream should return NextResponse.next() for browser requests. The adapter injects Link headers directly on the response object.
• fetchAsset should forward the optional RequestInit to fetch() so the subrequest header is included.

Options

| Option | Type | Default | Description | | ------------------ | ----------------------------------- | --------- | ----------------------------------------------------------- | | upstream | (req) => Response | — | Handler for non-bot requests (return NextResponse.next()) | | fetchAsset | (url, init?) => Response | — | Fetch a .md file by URL (forward init to fetch) | | trailingSlash | "never" \| "always" \| "preserve" | "never" | Trailing slash mode | | enableLinkHeader | boolean | true | Inject Link rel=alternate on HTML responses | | hooks | { onAIRequest?, onMiss? } | — | Lifecycle callbacks for AI request events | | redirects | { internal?, external? } | — | Redirect rules for AI bots | | skip | { prefixes?, extensions? } | — | Paths to skip entirely | | headers | { cacheControl? } | — | Custom response headers |

What it does

1. Subrequest detection to prevent fetch() infinite loops on Vercel
2. Trailing-slash enforcement (configurable: never, always, preserve)
3. AI bot detection via UA
4. Content negotiation via Accept header
5. Serves pre-built .md via fetchAsset with full AEO headers
6. Internal redirects: routes to target's .md
7. External redirects: returns markdown notice
8. 406 when neither HTML nor markdown is acceptable
9. Link header injection on upstream NextResponse.next() for HTML responses
10. Analytics hooks (onAIRequest / onMiss)

License

Apache 2.0