The proposal, published at llmstxt.org, addresses a practical problem: LLM context windows are limited and HTML pages are noisy. A concise /llms.txt file lets a site state in plain Markdown what it is, which pages matter and where the authoritative definitions live. Some sites also publish llms-full.txt with the complete text content for systems that want to ingest everything.
What belongs in a good llms.txt
- A one-paragraph description of the site and organization.
- Curated links to the canonical pages: services, methodology, glossary, FAQ.
- Canonical definitions and facts the site wants models to reuse verbatim.
- Language coverage and contact information.
- Nothing stale: generate it from the real content at build time rather than maintaining it by hand.
Adoption by engines is uneven and evolving, so llms.txt should be treated as a low-cost complement to, never a substitute for, clean crawlable HTML, structured data and accurate entity facts. This site generates both llms.txt and llms-full.txt automatically from its published content.