Problem

Developers increasingly rely on AI coding assistants to read and interpret documentation. If your documentation is only accessible via a web browser - cluttered with navigation elements, trackers, and complex HTML - AI agents consume excessive tokens on irrelevant data. This quickly exhausts their context windows.

Why it matters

Providing a clean, token-optimised text version of your documentation is the modern equivalent of providing a high-quality REST API. It ensures AI agents can quickly ingest your entire documentation set. This results in more accurate code suggestions and better support.

Approach

Use docmd’s built-in LLMs Plugin. This plugin natively implements the emerging llms.txt standard. It automatically generates token-optimised summaries and full-context files during every build process.

Implementation

Configure the llms plugin in your Plugin Configuration.

1. Configure the Site URL

Ensure the url property is correctly set in your docmd.config.json. This allows the plugin to generate absolute URLs for all pages in the llms.txt file.

{
  "title": "My Project Docs",
  "url": "https://docs.example.com",
  "plugins": {
    "llms": {}
  }
}

2. Output Files

During the build process, docmd generates two key files in your site root:

• llms.txt: A concise, structured Markdown summary of all your pages, including their titles, descriptions, and full URLs.
• llms-full.txt: A comprehensive file containing the raw Markdown content of your entire site, concatenated with standard separators (---). This provides the ultimate “source of truth” for AI models.

3. Controlling Ingestion

You can exclude specific pages from the AI-ready output using the llms property in the Page Frontmatter.

---
title: "Internal Debugging Guide"
llms: false
---

Trade-offs

Generating llms-full.txt creates a large single file. For exceptionally large documentation sites, this file could exceed several megabytes. While ideal for modern LLMs with large context windows (like Gemini 1.5 Pro or Claude 3.5 Sonnet), it may be too large for smaller models. Ensure you organise your Navigation logically so the AI can prioritise important sections.