Build an llms.txt file from a list of pages, or paste one to check its H1, links and sections. Follows the llmstxt.org proposal. Nothing is uploaded.
To make an llms.txt file, write a Markdown file with an H1 for the site name, a short blockquote summary and H2 sections of links, then serve it at /llms.txt. That is the layout in the proposal at llmstxt.org, and this page writes it for you. Only the H1 is required. After it come an optional blockquote, optional paragraphs or lists that hold no headings, and any number of H2 sections. Each section is a list where every item is a link, written as - [name](url), and may end with a colon and a note. Worked example: the Build sample lists 4 pages in 3 sections, Docs, Guides and Optional, and the file it makes passes the Check mode with no problems.
Write one page per line as Section | Title | URL | note. If you put a ## Section line above a group, you can drop the section from each line and write Title | URL | note. A line with only a URL is accepted too, and the title is taken from the last part of the address. Paths that start with / get the site address you type in front of them. A repeated address is listed once, a line with no address is skipped and its number is reported, and the section called Optional is moved to the end unless you turn that off. In the proposal, Optional is a convention for secondary links that an agent can skip when it needs a shorter context.
Check mode reads the file line by line and reports an error for a missing H1, an H1 that is not the first thing in the file, a heading other than the H1 and H2 before the first section, and a list item that is not a link, has no name, has an empty address or a space inside it, or has text after the link without a colon. It reports a warning for a second H1, a relative address, an address repeated on another line, a section with no list items, a repeated section name and plain text inside a section. The proposal does not forbid relative addresses, but full https:// addresses work best when a file is read away from your site. Each problem comes with its line number. In the sample, 5 problems appear, on lines 5, 7, 8, 10 and 12.
llms.txt is a proposal, not a standard, and no tool can say which AI products fetch it. This page checks the format only; it does not request your site, does not test that the links work, and does not claim any search or chat product reads the file. Treat the file as a tidy, human-readable index that an agent can use if it looks for it. Block rules for crawlers belong in robots.txt, which is a different file. Check mode stops listing after 500 problems, and the text you paste is processed in this tab only.
Type your site name and a one or two sentence summary, list your pages as Section | Title | URL | note, and copy the result. Save it as llms.txt and serve it at /llms.txt on your domain. With the sample you get 4 links in 3 sections.
Optional is the name the proposal uses for a section of secondary links. An agent that needs a shorter context can skip them. This page puts a section named Optional last and warns if you write it as optional with a small o.
The proposal starts the file with an H1 holding the project name, and it is the only required part. If a title line without a # or a paragraph comes before it, the report names the line where the other content starts, for example line 1.
This page cannot tell you. llms.txt is a proposal and the page does not claim that any named product reads it. Check each vendor's own documentation if that matters to you, and keep your robots.txt rules separate.
The format does not forbid it, so the checker gives a warning, not an error. Full addresses such as https://example.com/docs/start still work best because the file may be read without your site. Type the site address in Build mode and paths are expanded for you.
No. The text is processed in your browser and not uploaded. The mode and the options, including the site name and summary you type, are saved in this browser, but the page list and the file you check are not.