The @docmd/plugin-llms plugin follows the llms.txt standard. It generates two files at build time: a structured summary (llms.txt) and a full concatenated context (llms-full.txt). AI assistants and tools that understand the standard can use these to ingest your documentation directly.

The plugin is enabled by default in 0.8.8. To produce absolute links, set url in your docmd.config.json.

Generated Output

The plugin produces three files at the site root:

  • llms.txt — structured list of all pages with title + description + URL
  • llms-full.txt — the same list but with each page’s full markdown body concatenated below the entry
  • llms.json — a machine-readable manifest of every page (title, url, description, priority)

These files are linked in the page <head> for automatic discovery by AI tools.

Configuration

Option Type Default Description
enabled boolean true Enable or disable the LLM context generation.
fullContext boolean true If true, generates llms-full.txt containing the full markdown of every page.
maxTokenLimit number null Optional limit on the total characters/tokens for context files.
i18n boolean false When true, write per-locale files (llms.<locale>.txt, etc.) in addition to the default-locale set. See Multi-locale output below.

Example

docmd.config.json
{
  "url": "https://docs.example.com",
  "plugins": {
    "llms": {
      "fullContext": true
    }
  }
}

Default behaviour (0.8.8)

The plugin writes files for the default locale only. This is a deliberate change from earlier versions where the plugin wrote per-locale output by default. The reason: the unsuffixed llms.txt / llms-full.txt / llms.json filenames are the standard names that downstream consumers (Cursor, Claude, GPT, etc.) look for. Splitting them into llms.en.txt + llms.hi.txt + llms.fr.txt would have broken every existing integration.

For single-locale projects (no config.i18n block) this is invisible — the plugin writes a single set of files at the site root, same as before. For multi-locale projects, only the default-locale pages are in the bundle.

Multi-locale output (opt-in)

To get per-locale files, set i18n: true:

docmd.config.json
{
  "plugins": {
    "llms": { "i18n": true }
  }
}

The plugin then writes:

site/llms.txt          ← default locale (en) — UNSUFFIXED
site/llms-full.txt     ← default locale (en) — UNSUFFIXED
site/llms.json         ← default locale (en) — UNSUFFIXED
site/llms.ja.txt       ← Japanese — suffixed
site/llms-full.ja.txt  ← Japanese — suffixed
site/llms.ja.json      ← Japanese — suffixed
site/llms.fr.txt       ← French — suffixed
site/llms-full.fr.txt  ← French — suffixed
site/llms.fr.json      ← French — suffixed

Notice the pattern: the default locale never gets a suffix — its files keep the unsuffixed names so existing consumers don’t break. Only the non-default locales get a .<locale> suffix.

For sites with only one locale configured, no per-locale files are written regardless of the i18n flag (the suffix would just add noise).

Security: Sanitised Output (0.8.10)

Since 0.8.10, all user-controlled strings (page titles, descriptions) are sanitised before they land in llms.txt, llms-full.txt, and llms.json:

  • Markdown-special characters (`, [, ], newlines) in titles are escaped so a malicious frontmatter can’t break out of the [title](url) link form.
  • Strings starting with =, +, -, or @ are prefixed with a single-quote so opening the file in a spreadsheet (Excel / Sheets / LibreOffice) does not execute a formula.

The body in llms-full.txt is preserved verbatim — the file is treated as a first-party channel, and consumers that render it in an HTML context should sanitise on their side.

Excluding a Page

If a page contains sensitive information or internal notes you don’t want AI models to learn, use the llms: false flag in your frontmatter:

---
llms: false
---

This excludes the page from the LLMS files. The page is still rendered in the regular site HTML and is still included in search results.

See also

Maximising AI Accuracy

For detailed best practices on structuring your markdown (semantic headings, alt-text, etc.), see our Optimising for AI Agents guide.

Previous Analytics Next OpenAPI