Data substrate

llms.txt

A convention for publishing an LLM-friendly index of a website at /llms.txt — a curated markdown summary of what the site is and where its important content lives, written for AI crawlers and agents instead of browsers.

Operating principle

Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.

What it is

llms.txt, proposed by Jeremy Howard (Answer.AI) in 2024, is a markdown file at a site's root: a name, a short summary, and a curated list of links to the content that matters. An optional llms-full.txt carries expanded content. Documentation sites and AI-facing products adopted it quickly because it costs almost nothing to publish.

Why it matters

HTML built for browsers spends an agent's context window on navigation, scripts, and layout. A curated index lets an agent find the canonical content in one fetch. The convention is early and unevenly honored by crawlers — treat it as a cheap bet, one discovery surface among several.

How we use it

Product sites in our stack publish llms.txt alongside the other machine-readable surfaces — a hosted MCP server and Agent Skills indexes under /.well-known/ — so an agent can read what a product does and how to interact with it without scraping.

Related resources