llms-txt-kit 0.1.0

Generate and validate llms.txt files — CLI and Python library, zero dependencies.

llms-txt-kit

Generate and validate llms.txt files — CLI and Python library, zero dependencies, Python 3.8+.

llms.txt is a plain-text file at the root of your site that gives large language models a curated map of it: a title, a one-paragraph summary, and links to the pages that matter. Think robots.txt, but instead of where crawlers may go it says what your site means.

pip install llms-txt-kit

# build one from your sitemap, in one command
llms-txt generate --from-sitemap acme.coffee \
  --name "Acme Coffee" --description "Specialty roaster in Brooklyn." -o llms.txt

# check the one you already have
llms-txt validate acme.coffee

Prefer a web form? The hosted version is free at sempite.com/tools/llms-txt-generator — no install, no signup.

Why bother

AI assistants increasingly answer questions about businesses instead of linking to them. llms.txt is the one file where you get to state, in your own words and in a form models can parse, what your site is and which pages matter. It costs nothing to publish, and writing one forces a genuinely useful exercise: summarising your business in a paragraph a machine can quote.

CLI

generate

# explicit links
llms-txt generate \
  --name "Acme Coffee" \
  --description "Specialty coffee roaster in Brooklyn, shipping nationwide." \
  --url acme.coffee \
  --link "Menu|/menu|What we're pouring this week" \
  --link "Wholesale|/wholesale|Bulk and cafe accounts" \
  --optional "Press kit|/press" \
  -o llms.txt

# or derive everything from the sitemap
llms-txt generate --from-sitemap acme.coffee \
  --name "Acme Coffee" --description "Specialty roaster." \
  --limit 25 --titles

--from-sitemap reads sitemap.xml (following sitemap indexes), groups URLs into sections by their first path segment, folds one-off segments into Pages so you don't get a heading per page, and routes boilerplate — privacy, terms, accessibility — into the conventional ## Optional section. --titles fetches each page for its real <title> and meta description instead of deriving names from slugs.

Flag	Meaning
--name / --description	required: the H1 and the > summary
--url	canonical site; also absolutises relative --link URLs
--link "Title|URL|note"	a key page — repeatable
--optional "Title|URL|note"	a page for ## Optional — repeatable
--from-sitemap URL	build from a sitemap instead of --link flags
--limit / --titles	cap link count; fetch real titles
--no-credit	omit the trailing credit comment
-o FILE	write to a file instead of stdout

validate

llms-txt validate acme.coffee        # fetches https://acme.coffee/llms.txt
llms-txt validate ./llms.txt         # local file
llms-txt validate acme.coffee --json # machine-readable

https://acme.coffee/llms.txt
----------------------------
  PASS  llms.txt is reachable  —  https://acme.coffee/llms.txt
  PASS  Starts with an H1 title  —  # Acme Coffee
  PASS  Exactly one H1  —  One H1
  PASS  Has a blockquote summary  —  Specialty coffee roaster in Brooklyn…
  PASS  Links to key pages  —  7 link(s)
  WARN  Links have descriptions  —  2/7 links annotated
  PASS  Reasonable size (under 100 KB)  —  4 KB

12/13 checks passed · 0 error(s), 1 warning(s)

Exit codes make it CI-friendly: 0 valid, 1 errors found (--strict also fails on warnings), 2 bad usage or unreachable target.

# keep llms.txt honest on every deploy
- run: pipx run llms-txt-kit validate ./public/llms.txt --strict

Python API

from llmstxt import generate, parse, validate, validate_url, from_sitemap

text = generate(
    "Acme Coffee",
    "Specialty coffee roaster in Brooklyn, shipping nationwide.",
    links=[("Menu", "https://acme.coffee/menu", "What we pour")],
    optional_links=[("Press kit", "https://acme.coffee/press")],
    url="https://acme.coffee",
)

report = validate_url("acme.coffee")
print(report.ok, len(report.errors), len(report.warnings))
for check in report.checks:
    print(check.status, check.label, check.detail)

doc = parse(text)
print(doc.title, doc.summary, [l.url for l in doc.links])

text = from_sitemap("acme.coffee", "Acme Coffee", "Specialty roaster.", limit=25)

generate() accepts links as Link objects, (title, url, note) tuples, or dicts — whichever your data already looks like. validate() never raises on malformed input; it reports.

What validation checks

Errors (the file is not spec-shaped): reachable, non-empty, not HTML, starts with a single # H1, has at least one link, all links absolute. Warnings (it works, but could be better): blockquote summary present, three or more links, HTTPS everywhere, no duplicate URLs, links annotated with notes, sections used, under 100 KB.

Where the file goes

Serve it at the root of your domain as plain text — https://yoursite.com/llms.txt — exactly like robots.txt. Nginx:

location = /llms.txt { default_type text/plain; }

Related

• The specification: llmstxt.org
• The spec authors' own Python package: llms-txt (parsing/expansion oriented) — this project is unaffiliated and complementary, focused on generating and linting the file itself
• Hosted, no-install version of this tool: sempite.com/tools/llms-txt-generator
• Can AI actually read your site? Free AI Readiness Score — 21 technical checks
• Does AI cite you? Free AI Citation Checker

Development

git clone https://github.com/SEMPITEHQ/llms-txt-kit
cd llms-txt-kit
python -m unittest discover -s tests -v

No dependencies, no build step, no test framework to install.

License

MIT © SEMPITE