llms.txt Spec

llms.txt is a machine-readable index of documentation, designed for AI agents to discover and consume content efficiently. Eve Horizon generates this file automatically at build time, making the entire docs surface available for agent consumption without scraping or manual indexing.

What is llms.txt?

The llms.txt specification defines a standard way for websites to expose their content to large language models. It is the equivalent of robots.txt for AI agents: a well-known file at a predictable URL that describes what content is available and where to find it.

Two files are served:

File	Content	Use case
/llms.txt	Title, one-line summary, and URL per page	Agent discovery and navigation
/llms-full.txt	Full rendered text of all pages, stripped of HTML	Deep content ingestion

The lightweight llms.txt is what agents check first. It provides enough context to decide which pages are relevant. The full-text variant (llms-full.txt) is available when an agent needs to ingest the entire documentation set.

How Eve Horizon uses it

Eve Horizon's docs site auto-generates both files from the Docusaurus source at build time. Every documentation page in the site is included in the index. Blog posts (the changelog) and standalone pages are excluded by configuration.

The result: any agent that knows the docs site URL can fetch /llms.txt and immediately understand what documentation is available, what each page covers, and where to find the full content.

Format example

# Eve Horizon

> Ship software with AI-powered workflows

## Get Started

- [What is Eve Horizon?](https://docs.eve-horizon.dev/docs/get-started/what-is-eve-horizon): A job-first platform for running AI-powered skills against Git repositories.
- [Quickstart](https://docs.eve-horizon.dev/docs/get-started/quickstart): Zero to deployed in five minutes.

## Agent Integration

- [Agent-Native Design](https://docs.eve-horizon.dev/docs/agent-integration/agent-native-design): Design principles for building applications where agents are first-class citizens.
- [Skills System](https://docs.eve-horizon.dev/docs/agent-integration/skills-system): Architecture deep dive into the skills system.

Each entry has a title (from the page's frontmatter), a URL, and a description (from the page's description frontmatter field). Sections are grouped by the sidebar hierarchy.

Auto-generation from Docusaurus

The llms.txt and llms-full.txt files are generated by the @signalwire/docusaurus-plugin-llms-txt plugin. The plugin runs at build time and produces static files that are served alongside the rest of the site.

Configuration

The plugin is configured in docusaurus.config.ts:

plugins: [
  [
    '@signalwire/docusaurus-plugin-llms-txt',
    {
      content: {
        enableMarkdownFiles: true,
        enableLlmsFullTxt: true,
        includeBlog: false,
        includePages: false,
        includeDocs: true,
      },
    },
  ],
],

Option	Value	Description
enableMarkdownFiles	true	Generate the index from markdown sources
enableLlmsFullTxt	true	Also generate the full-text variant
includeBlog	false	Exclude changelog/blog posts
includePages	false	Exclude standalone pages (homepage, etc.)
includeDocs	true	Include all docs pages

note

The options go under the content key. This is specific to the @signalwire/docusaurus-plugin-llms-txt plugin (v1.2.2).

Build output

After running pnpm build, the generated files are available in the build output directory:

build/
├── llms.txt
├── llms-full.txt
└── ...

Both files are served as static assets at the site root.

Accessing the endpoints

Once deployed, the files are accessible at:

URL	Content
https://docs.eve-horizon.dev/llms.txt	Page index with titles, summaries, and URLs
https://docs.eve-horizon.dev/llms-full.txt	Full rendered text of all documentation

During local development (pnpm start), the files are not generated because the dev server does not run the full build pipeline. To test llms.txt locally, use pnpm build && pnpm serve.

How agents discover and use it

Agents discover the docs through llms.txt in several ways.

Direct discovery

An agent that knows the docs site URL can fetch /llms.txt to get a complete index of available documentation. This is the simplest path and works for any agent with HTTP access.

Skill-based discovery

The eve-read-eve-docs skill in the eve-work pack teaches agents how to find and use the Eve documentation. The skill references the docs site and provides instructions for navigating the content hierarchy.

Eve CLI integration

Agents running inside Eve jobs have access to the CLI and can discover documentation through the skills system. The skill read eve-read-eve-docs command provides navigation instructions that include the docs site URL.

Agent workflow

A typical agent workflow for consuming documentation:

1. Fetch /llms.txt to get the page index.
2. Scan titles and descriptions to identify relevant pages.
3. Fetch the specific page URL for detailed content.
4. Alternatively, fetch /llms-full.txt if the full corpus is needed.

The lightweight index is designed to fit comfortably within an agent's context window, enabling efficient navigation without loading all content upfront.

Why it matters

The llms.txt standard solves a practical problem: agents need to discover what documentation exists and what it covers before they can use it effectively. Without a machine-readable index, agents must either be pre-loaded with documentation knowledge (which drifts) or scrape HTML pages (which is fragile and expensive).

By generating llms.txt automatically from the same source that produces the human-readable docs, the index stays in sync with the content. When a new page is added or a description is updated, the next build produces an updated index. No manual maintenance required.