AI & LLMs

Docs for LLM

You can make your docs site more AI-friendly with dedicated docs content for large language models.

To begin, make a getLLMText function that converts pages into static MDX content.

In Fumadocs MDX, you can do:

lib/get-llm-text.ts

import { source } from '@/lib/source';
import type { InferPageType } from 'fumadocs-core/source';

export async function getLLMText(page: InferPageType<typeof source>) {
  const processed = await page.data.getText('processed');

  return `# ${page.data.title} (${page.url})

${processed}`;
}

It requires includeProcessedMarkdown to be enabled:

source.config.ts

import { defineDocs } from 'fumadocs-mdx/config';

export const docs = defineDocs({
  docs: {
    postprocess: {
      includeProcessedMarkdown: true,
    },
  },
});

`llms-full.txt`

A version of docs for AIs to read.

app/llms-full.txt/route.ts

import { source } from '@/lib/source';
import { getLLMText } from '@/lib/get-llm-text';

// cached forever
export const revalidate = false;

export async function GET() {
  const scan = source.getPages().map(getLLMText);
  const scanned = await Promise.all(scan);

  return new Response(scanned.join('\n\n'));
}

`*.mdx`

Allow AI agents to get the content of a page as Markdown/MDX, by appending .mdx to the end of path.

Make a route handler to return page content, and a middleware to point to it:

app/llms.mdx/[[...slug]]/route.ts

import { getLLMText } from '@/lib/get-llm-text';
import { source } from '@/lib/source';
import { notFound } from 'next/navigation';

export const revalidate = false;

export async function GET(
  _req: Request,
  { params }: RouteContext<'/llms.mdx/[[...slug]]'>,
) {
  const { slug } = await params;
  const page = source.getPage(slug);
  if (!page) notFound();

  return new Response(await getLLMText(page), {
    headers: {
      'Content-Type': 'text/markdown',
    },
  });
}

export function generateStaticParams() {
  return source.generateParams();
}

next.config.ts

import type { NextConfig } from 'next';

const config: NextConfig = {
  async rewrites() {
    return [
      {
        source: '/docs/:path*.mdx',
        destination: '/llms.mdx/:path*',
      },
    ];
  },
};

`Accept`

To serve the Markdown content instead for AI agents, you can leverage the Accept header.

proxy.ts (Next.js)

import { NextRequest, NextResponse } from 'next/server';
import { isMarkdownPreferred, rewritePath } from 'fumadocs-core/negotiation';

const { rewrite: rewriteLLM } = rewritePath('/docs/*path', '/llms.mdx/*path');

export default function proxy(request: NextRequest) {
  if (isMarkdownPreferred(request)) {
    const result = rewriteLLM(request.nextUrl.pathname);

    if (result) {
      return NextResponse.rewrite(new URL(result, request.nextUrl));
    }
  }

  return NextResponse.next();
}

Page Actions

Common page actions for AI, require *.mdx to be implemented first.

AI Page Actions

npx @fumadocs/cli add ai/page-actions

Use it in your docs page like:

app/docs/[[...slug]]/page.tsx

<div className="flex flex-row gap-2 items-center border-b pt-2 pb-6">
  <LLMCopyButton markdownUrl={`${page.url}.mdx`} />
  <ViewOptions
    markdownUrl={`${page.url}.mdx`}
    githubUrl={`https://github.com/${owner}/${repo}/blob/dev/apps/docs/content/docs/${page.path}`}
  />
</div>

Ask AI

You can install the AI search dialog using Fumadocs CLI:

npx @fumadocs/cli add ai/search

Then, add the <AISearchTrigger /> component to your root layout.

AI Model

By default, it's configured for Inkeep AI using Vercel AI SDK.

To setup for Inkeep AI:

Add your Inkeep API key to environment variables:
```
INKEEP_API_KEY="..."
```

Add the AISearchTrigger component to root layout (or anywhere you prefer):

import { AISearchTrigger } from '@/components/search';

export default function RootLayout({
  children,
}: {
  children: React.ReactNode;
}) {
  return (
    <html lang="en">
      <body>
        <AISearchTrigger />
        {children}
      </body>
    </html>
  );
}

To use your own AI models, update the configurations in useChat and /api/chat route.

Note that Fumadocs doesn't provide the AI model, it's up to you. Your AI model can use the llms-full.txt file generated above, or more diversified sources of information when combined with 3rd party solutions.

Docs for LLM

llms-full.txt

*.mdx