hexo-generator-llms
v0.1.1
Published
GEO for Hexo: generate a Markdown (.md) twin of every page plus llms.txt so AI/LLM crawlers get clean, token-efficient content (generative engine optimization).
Maintainers
ssarcandyReadme
hexo-generator-llms
GEO (Generative Engine Optimization) for Hexo.
Generate a clean Markdown (.md) copy of every Hexo page for AI/LLM crawlers, advertise it
with <link rel="alternate" type="text/markdown">, and emit an llms.txt index — the
static-site equivalent of Cloudflare's "serve Markdown to AI" content-negotiation, done at
build time (works on GitHub Pages and any static host).
Why: LLMs pay per token, and a page of <div>/class/script HTML costs ~4–5× the tokens of
the same content in Markdown. Serving a clean .md saves the crawler tokens and helps it read
your content hierarchy (headings, lists, tables) without visual-markup noise.
Install
npm install hexo-generator-llmsThe plugin auto-loads (its name starts with hexo-). On the next hexo generate it emits, for
each page, an index.md next to the index.html:
public/2016/12/25/my-post/index.html
public/2016/12/25/my-post/index.md ← new
public/llms.txt ← newand injects into each page's <head>:
<link rel="alternate" type="text/markdown" href="https://example.com/2016/12/25/my-post/index.md">How it works
- Converts the rendered page HTML (
post.content) to Markdown with Turndown +turndown-plugin-gfm(tables, fenced code, strikethrough). Using rendered HTML means any templating tags in your source Markdown are already expanded — nothing leaks into the.md. - Rewrites image/link URLs to absolute so a fetched
.mdis self-contained. - Records which pages got a
.mdand injects thealternatelink only on those, by reading each page's own<link rel="canonical">inside anafter_render:htmlfilter — no theme edits required. - Only emits for pages that resolve to an HTML document (path ends in
/or.htm(l)). Files Hexo copies verbatim but still lists as pages —sw.js,web-app-manifest.json, a rawfeed.xml, any data JSON in your source root — are skipped automatically: converting them to Markdown is meaningless, and emitting<asset>/index.mdwould make Hexo create a directory that collides with the asset file of the same name.
Themes without a canonical link
The alternate link injection keys off <link rel="canonical">. If your theme doesn't emit
one, add this one line to your <head> instead:
<%- md_alternate_link() %>Configuration
All optional — sensible defaults, works with no config:
# _config.yml
llms:
enable: true
types: [post, page] # which sources to emit a .md for
exclude: [] # paths to skip, e.g. [projects, photography]
llms_txt: true # /llms.txt index
llms_full_txt: false # /llms-full.txt (all bodies concatenated)
alternate_link: true # inject <link rel="alternate"> into <head>
front_matter: true # YAML header (title/source/date/tags) in each .md
absolute_urls: true # rewrite links/images to absolute URLsData-driven pages (e.g. a portfolio or gallery whose content is rendered from a template/data
file rather than a Markdown body) have no meaningful body to convert — list their paths in
exclude to skip them.
Note on GitHub Pages
GitHub Pages can't set a custom Content-Type on .md files, but AI crawlers discover the
Markdown via the rel="alternate" hint, the .md extension, and llms.txt — not the response
MIME type — so it works regardless.
License
MIT