H1 with project name + one-line blockquote summary.
llms.txt examples
llms.txt examples and patterns.
Real llms.txt files from sites that ship one in 2026 — what's in each, the patterns worth copying, and where the spec leaves room for opinionated choices.
Completely free for 30 days. No credit card required.
01Details
What a minimal llms.txt looks like
Two or three sections (Docs, API reference, Examples) with Markdown links to the .md version of each page.
That is it. No required JSON, no manifest schema, no version field.
02Details
Patterns that work in practice
Group sections by reader intent (Getting started / API reference / Examples / Migration), not by site IA.
Always link to the .md file, not the .html one. Agents that follow links get clean text instead of nav chrome.
Keep the file under 50 KB. If it gets longer, move the long tail under an Optional section.
Put authoritative content first. Tutorials before deep dives, getting-started before edge cases.
03Details
Anti-patterns to avoid
Linking to HTML pages with no Markdown equivalent. Agents that fetch them waste tokens on layout markup.
Mixing canonical sources with marketing pages. The Optional section is the place for landing pages, not the main index.
Dumping the entire sitemap. llms.txt is curation, not enumeration — that is what sitemap.xml is for.
Hand-written files that go stale. If your docs change weekly and your llms.txt does not, agents will trust the wrong things.
04Details
Concrete examples worth opening side by side
Anthropic's developer docs at docs.anthropic.com/llms.txt — large, well-grouped, links every reference page to its .md counterpart. The reference implementation for an API doc surface.
FastHTML at fastht.ml/llms.txt — small, framework-focused, shows the minimum-viable shape (H1 + blockquote + a single section of links).
Cursor at cursor.com/llms.txt — short, product-focused, demonstrates the Optional section pattern.
Mintlify-hosted sites (e.g. mintlify.com/docs/llms.txt) — automatic emission, so most Mintlify-built docs have one by default.
Bloom-generated llms.txt (in preview) — emitted from OpenAPI alongside the SDK pipeline; structurally consistent because it's deterministic from the spec.
05Details
Category shapes you'll see in the wild
API tool docs (Stripe-style references with operation-grouped sections).
AI-first product docs (LangChain, Anthropic-style developer hubs).
Open-source library docs (per-package READMEs aggregated by topic).
Generated docs from OpenAPI tools (Bloom, Mintlify, Fern, ReadMe) — most structurally consistent because they emit deterministically.
06Details
How to write yours by copying the best patterns
Start from the spec at llmstxt.org — keep the H1 / blockquote / sections / Optional shape.
Audit your existing docs site for canonical sources worth surfacing.
Ship per-page Markdown alongside HTML so the links resolve to text agents can consume.
Automate the regeneration. The single biggest reason llms.txt files go stale is a human writing them.