llms.txt Generator

/llms.txt · markdown · built and validated as you type

robots.txt tells a crawler what it may not fetch. llms.txt tells a language model what is worth fetching: your title, one line about what this site is, and a curated list of pages with a sentence each. It is markdown with two rules that are easy to break — exactly one H1 at the top, and no heading deeper than H2 — so this builds it from whatever shape your URL list is in, and checks it.

What llms.txt is

A markdown file at the root of your site — https://example.com/llms.txt — proposed in 2024 as a way to hand a language model a short, curated map of a site instead of leaving it to crawl the navigation, the cookie banner and the footer. The format is deliberately tiny:

# Project name

> One-sentence summary of what this is.

Any amount of ordinary markdown, with no headings in it.

## Docs

- [Quickstart](https://example.com/docs/quickstart): five minutes to a first request
- [API reference](https://example.com/docs/api)

## Optional

- [Blog](https://example.com/blog)

The H1 is the only required element. Everything else is optional, and the sections are just H2 headings with a bullet list of markdown links under them, each with an optional description after a colon.

The two rules people break

• Exactly one H1, and it comes first. A file that opens with a comment, a badge line or a table of contents fails the first thing a parser checks.
• No heading below H2. An H3 inside a section does not create a sub-section — for a consumer walking H2 to H2, it is unstructured text in the middle of a link list. This flags it as a problem rather than a style note.
• Links must be absolute. The file is read away from your server, often by something that never saw the page it came from. /docs/api resolves to nothing there. Give the site URL above and relative links are expanded for you.

Paste your links in whatever shape you have them

The links box accepts a mix, one entry per line, and works out which is which:

• ## Section name — starts a new section. Everything after it goes under that heading.
• - [Title](https://url): description — already markdown, used as-is.
• https://example.com/docs/api: the reference — the description follows the colon.
• API reference — https://example.com/docs/api — title first, separated by a dash, comma or pipe.
• https://example.com/docs/getting-started — no title at all, so one is made from the last path segment: Getting started.

Lines with no URL in them are skipped and reported, rather than being emitted as broken bullets.

The "Optional" section

## Optional is the one section name with a defined meaning: links under it may be skipped when a shorter context is needed. Everything else is a name you choose. If you have several sections and none of them is Optional, this says so — deciding what a model can afford to skip is the part of the file that does the work.

What gets checked

• A single H1, first in the file; no H3 or deeper anywhere.
• Every bullet is a real markdown link with title text.
• Every URL is absolute; http:// is flagged, and can be rewritten to https://.
• Duplicate links across all sections — dropped or reported, your choice.
• Empty sections, a missing summary, and a details block that smuggled in a heading.
• Tracking parameters (utm_*, fbclid, gclid and the rest) stripped, because a curated map should not be carrying campaign tags.

Publishing it

1. Serve the file at /llms.txt — the root, not /.well-known/, which is where security.txt goes.
2. Send it as text/markdown; charset=utf-8, or text/plain if your host will not do the former. Both are read fine; neither should be text/html.
3. Keep it small. It is an index, not a corpus: a link with a good sentence beats ten links without one.
4. The related llms-full.txt convention puts the actual content of those pages in one file. That is a different job — this tool builds the index.

FAQ

Related URL tools