README

Generate an /llms.txt index of your site and clean per-page Markdown (.md) versions of pages, so LLMs / AI agents can read your content efficiently. No static-publish-queue dependency — everything is generated on the fly and cached.

Features

/llms.txt — a spec-shaped Markdown index: # Site name, a > tagline blockquote, a ## Pages link list and an optional ## Optional section for secondary (non-menu) pages. Each entry is - [Title](url): short summary.
Per-page Markdown — request …/a-page.md or send Accept: text/markdown to any page URL to get a clean Markdown rendering (Title + meta description + body, including Elemental block content).
Summaries fall back MetaDescription → Content field → Elemental blocks (getElementsForSearch()), word-limited to keep the index concise.
Catalogue DataObjects — also list DataObjects that live under a page (products, caravan models, listings…) in the index with their own URL + summary, via dataobject_classes. Each record needs a Title and a Link()/AbsoluteLink().
Per-record Markdown — a page controller can expose a detail record with getLLMsRecord(): ?DataObject, so …/<action>/<slug>.md returns that record's Markdown (Title + summary + Content/Description body), not the page's.
Extension hooks — updateLLMsMarkdown(&$md) (append structured Markdown to a page or record) and getLLMsSummary() (a one-line index summary; return false to exclude the record).
<head> discovery tags — every page advertises its Markdown alternate (the current URL + .md) and the index (<link rel="describedby">).
Cached per locale (index) and per subject + LastEdited (Markdown), with HTTP Cache-Control; cleared on ?flush and on page publish/unpublish.
Scales: max_pages, max_depth, menu_only, exclude_classes, metadescription_only_classes, dataobject_max, and a canView() guard.

Requirements

Silverstripe CMS ^6
league/html-to-markdown ^5.1 (installed automatically)

Installation

composer require xddesigners/silverstripe-llms-txt
dev/build flush=all

Configuration

XD\LLMsTxt\Controllers\LLMsTxtController:
  cache_ttl: 3600         # seconds to cache output (0 = off; also drives .md cache + Cache-Control)
  optional_section: true  # split non-menu pages into "## Optional"
  max_pages: 300          # cap listed pages (0 = unlimited)
  max_depth: 2            # only pages up to N levels deep (0 = unlimited, 1 = top-level)
  menu_only: false        # only list pages shown in menus
  exclude_classes:        # page classes to leave out
    - Path\To\SomePage
  metadescription_only_classes:  # summarise these from MetaDescription only (matched via instanceof)
    - Path\To\ListingPage
  dataobject_classes:     # DataObjects to also list, as [ClassName: 'Section heading']
    'App\Model\Product': 'Products'
  dataobject_max: 0       # cap records per DataObject class (0 = unlimited)

Extending

Add richer output from your own page types and models:

updateLLMsMarkdown(&$md) — on a page/record (or an extension of it), append structured Markdown after the title. Good for a product's price/brand/SKU, or a listing page enumerating its child DataObjects.
getLLMsSummary(): string|false — on a DataObject listed via dataobject_classes, return its one-line index summary. Return false to exclude the record (e.g. sold/inactive items).
getLLMsRecord(): ?DataObject — on a page controller, return the detail record currently being viewed (resolved from the request), so …/<action>/<slug>.md renders that record instead of the page. It runs at onAfterInit, before action params are set, so parse the URL rather than reading param().
getResolvedMetaDescription(): string — on a page, return the description the site actually renders (with your own fallbacks, e.g. content → homepage → site default). When present it's used for the index summary and the .md description, so llms.txt matches your <meta name="description">.

Records rendered as .md use their Content field, else Description.

License

BSD-3-Clause. See LICENSE.

xddesigners / silverstripe-llms-txt

Package info

Security

Statistics