Skip to content

llms.txt for AI tools

llms.txt is a convention: a Markdown file at a site’s root that tells AI tools what a project is and where to read more. LANrurugi’s homepage and this documentation site both publish them. Give an AI assistant the right address and it can answer questions, help you deploy, write scripts against the API or write plugins from the documentation, instead of guessing from whatever it saw in training.

Address Contents Size (approx.)
lrrg.dev/llms.txt Entry point: what the project is, the HTTP API and the plugin SDK, and links to the main docs 3.5 KB
docs.lrrg.dev/llms.txt Index of this site: links to the files below, plus the API and SDK overview 3 KB
docs.lrrg.dev/llms-full.txt Every documentation page, in both languages, as one Markdown file 800 KB
docs.lrrg.dev/llms-small.txt Abridged: guides and plugin authoring only, without the API pages and asides 80 KB
…/_llms-txt/guides-english.txt Guides and plugin authoring, English only 40 KB
…/_llms-txt/guides-simplified-chinese.txt The same, Simplified Chinese 40 KB
…/_llms-txt/api-reference-english.txt The API reference, English only, one section per group 300 KB
…/_llms-txt/api-reference-simplified-chinese.txt The same, Simplified Chinese 300 KB

Two structured files are linked from the llms.txt files as well:

  • docs.lrrg.dev/openapi.json: the complete HTTP API as OpenAPI 3.1. It is generated from the server’s own route registrations, so it lists exactly the endpoints the server serves.
  • sdk.lrrg.dev/plugin-sdk.json: the plugin SDK reference as JSON, generated by deno doc from the TypeScript the plugin runtime loads.

All of them are regenerated with every documentation release, from the same source as the pages, so they never drift apart.

You want the AI to Give it
Explain the project, answer general questions lrrg.dev/llms.txt
Help you deploy, debug an install or a reverse proxy One language’s guides (guides-*.txt) or llms-small.txt
Write a script or client for the API openapi.json: more precise than the API pages’ Markdown, and fewer tokens
Write a plugin Plugin authoring plus plugin-sdk.json
Have all the background, with a large enough context llms-full.txt

If you only need a metadata, login or download plugin for one site, the instance’s own AI Plugin Creation Wizard fetches the target pages, writes the code and trial-runs it, which is usually less work than having an outside assistant start from nothing.

  • AI tools that can browse (chat assistants with web reading, coding agents): send the address, for example “Read https://lrrg.dev/llms.txt first, then …”. It follows the links inside for the parts it needs.
  • Tools without web access: download the files you need and attach or paste them.
  • Coding agents working in a repository: download openapi.json or plugin-sdk.json into the project, which is steadier than fetching it every time.