llms.txt generator

An llms.txt file is a Markdown file you place at the root of your site, served at /llms.txt, that tells large language models what your project is and which pages are worth reading. It exists because an assistant answering a question about your product has to guess which of your pages to read, and it reads the raw HTML it is served, not the rendered experience a person sees. A short, curated map at a predictable location removes the guesswork.

The format is deliberately small. The specification at llmstxt.org defines a single required element and a few optional ones. The one required element is an H1 heading with the name of your project or site. Directly after it, a blockquote holds a short summary that carries the key information needed to understand the rest of the file. After the summary you may add zero or more Markdown sections of any kind except headings: paragraphs or lists that give more detail. Then come the link sections.

Each link section is an H2 heading followed by a Markdown list. Every list item is a link written as a bulleted Markdown hyperlink, optionally followed by a colon and a short note. The shape is exactly this: a dash, the link text in square brackets, the URL in parentheses, then an optional colon and notes. Group your pages into sections that make sense to a reader, such as Docs, Guides, or API, and keep each list focused on the pages that actually answer questions.

One section name is special. An H2 titled Optional marks links that an assistant can skip when it needs a shorter context. Put secondary material there, such as changelogs or deep reference pages, so the primary sections stay small enough to fit inside a limited context window. Everything outside the Optional section is treated as core.

Here is a minimal, valid example. The H1 names the project, the blockquote summarizes it, and two H2 sections list the pages, with an Optional section at the end for material an agent may skip:

# Example Docs

> Documentation for Example, a tool that does one thing well. Start with the quickstart, then the guides.

## Docs

- [Quickstart](https://example.com/docs/quickstart): Install and run in five minutes
- [Configuration](https://example.com/docs/config): Every setting explained

## Optional

- [Changelog](https://example.com/changelog): Release history

The generator above builds this for you. In From URL mode it reads your sitemap.xml, falls back to the Sitemap lines in your robots.txt when there is no sitemap at the usual path, and pulls your homepage title and description. It groups the URLs it finds by their first path segment into sections and fills in a draft. You then edit everything by hand: rename sections, reorder and rewrite the links, add notes, and move anything secondary into the Optional section. The preview on the right is always valid llms.txt, so what you copy or download is what you publish.

Once the file reads well, save it as llms.txt and serve it at your site root so it resolves at /llms.txt. Many teams also publish an llms-full.txt, a companion file that inlines the full Markdown of each page rather than linking to it, for assistants that prefer one document over a set of URLs. The generator focuses on the index file, which is the one the specification defines and the one most agents look for first.

A generated file is a snapshot of your site at one moment. The harder part is keeping it accurate as you add and rename pages, because a stale llms.txt points assistants at content that moved or no longer exists. The durable fix is to generate the file from your live content on every publish rather than editing a static copy, which is how Topicary produces llms.txt and per-page Markdown for the sites it publishes.