What is the llms.txt format?

llms.txt is a Markdown resource map that can live at the root or a more specific path. The August 2026 proposal also defines HTML and HTTP Link relations for discovery.

What sections are required in llms.txt?

Only an H1 heading is required. A blockquote summary, explanatory preamble, H2 file-list sections, link notes and an Optional-labelled section are optional.

What is the difference between llms.txt and llms-full.txt?

llms.txt is an index file with links and short descriptions. llms-full.txt contains the complete content of those linked documents concatenated together, it is meant for AI systems that need the full text in one request.

What Markdown is allowed in llms.txt?

The file starts with one H1. An optional blockquote and heading-free Markdown preamble may follow. Optional H2 sections contain Markdown list links with optional notes.

How large should llms.txt be?

The proposal defines no byte limit. Keep the map task-focused, measure it in your own clients, and split real content scopes into path-level files when useful.

llms.txt Format, Spec Reference

The complete reference for the llms.txt file format: required and optional elements, Markdown rules, link syntax, the llms-full.txt variant, and a checklist to validate your file.

Last updated: August 12, 2026

Format overview

llms.txt is a plain-text resource using CommonMark Markdown. It may live at the origin root or a more specific path such as https://example.com/docs/llms.txt. The most specific applicable file describes that scope.

Compatible coding assistants, RAG pipelines and research agents can use the file as a curated map. Publication alone does not prove discovery or use, so declare it with rel="describedby" where appropriate and measure the clients you actually support.

Required elements

The August 2026 proposal defines one mandatory element:

H1 heading, the first line must be an H1 (# Name) containing the project or site name.

The blockquote summary, explanatory preamble, H2 file lists and link notes are optional.

# Your Project Name

> One sentence describing what your project does and who it's for.

## Documentation

- [Getting Started](https://example.com/docs/start): Install and first steps.
- [API Reference](https://example.com/api): Full endpoint catalog.

## Optional

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

Optional sections

Sections are H2 headings (##) followed by unordered lists of Markdown links. Common section names include:

Documentation, primary docs, guides, references.
Product, marketing pages, pricing, status.
Examples, code samples, tutorials, demos.
Optional, changelog, blog, GitHub, lower priority for AI.
API, dedicated API reference section.
SDKs, language-specific client libraries.

Optional remains a clear editorial label for secondary resources, but v2 assigns it no special processing semantics.

# Acme SaaS

> Acme SaaS helps teams automate their billing workflows with a no-code dashboard
> and a REST API supporting 40+ payment providers.

## Product

- [Overview](https://acme.com/product): Core capabilities and use cases.
- [Pricing](https://acme.com/pricing): Plans, limits, and enterprise options.
- [Status](https://status.acme.com/): Uptime and incident history.

## Documentation

- [Quickstart](https://acme.com/docs/quickstart): Set up in under 5 minutes.
- [API Reference](https://acme.com/docs/api): REST endpoints, auth, rate limits.
- [SDKs](https://acme.com/docs/sdks): Node, Python, Ruby, Go clients.
- [Webhooks](https://acme.com/docs/webhooks): Event payloads and retry policy.

## Examples

- [Node.js integration](https://acme.com/examples/node): End-to-end payment flow.
- [Python integration](https://acme.com/examples/python): Subscription management.

## Optional

- [Changelog](https://acme.com/changelog): Version history.
- [Blog](https://acme.com/blog): Product updates and tutorials.
- [GitHub](https://github.com/acme/acme-oss): Open-source components.

Link syntax

Each link follows the Markdown syntax - [Title](URL): short description.

Use absolute URLs including the scheme (https://).
The description after the colon is plain text, keep it under ~120 characters and make it informative for an AI, not keyword-stuffed.
One link per list item, do not nest bullets.
Prefer canonical URLs (with trailing slash if that's your convention).

llms-full.txt variant

The wider ecosystem uses an optional companion file at /llms-full.txt. While llms.txt is an index of links, llms-full.txt contains the full text of those linked pages concatenated together, formatted in Markdown.

A tool that explicitly supports the convention may fetch llms-full.txt for consolidated context. The trade-off is greater size, staleness and security exposure.

Read the dedicated llms-full.txt guide for generation strategies.

Checklist

File served at the root or the intended scoped path
Plain-text or Markdown media type with UTF-8 encoding
Starts with exactly one H1 heading
Optional blockquote and preamble contain no headings
Link targets resolve correctly in their publication context
Optional link notes are concise and factual
File size is justified by the tasks and tested clients
No HTML tags, no nested lists
Validated with the llms.txt validator

Common mistakes

Relative URLs, - [Docs](/docs) will not resolve correctly when the file is fetched by an AI crawler. Always use absolute URLs.
Missing blockquote, skipping the > summary makes the file harder for models to parse and is non-compliant with the spec.
Wrong Content-Type, serving with text/html or no content-type causes some parsers to reject the file.
Keyword stuffing descriptions, AI models read these literally. Stuffing keywords degrades the quality signal.
Listing every page, curate your most important 10–30 links. An exhaustive sitemap belongs in llms-full.txt, not the index.

Related guides

How to create llms.txt, step-by-step for any stack.
llms-full.txt guide, full-content companion file.
Validator, check your file for spec compliance.
Generator, create a file from a form.
Best practices, what to include and what to skip.
Which AI crawlers read llms.txt, adoption status.