# For AI agents (/docs/ai) These docs are meant to be read by agents as much as by people. Three things to know: ## llms.txt and llms-full.txt [#llmstxt-and-llms-fulltxt] * **`/llms.txt`** — an index of every page, with a one-line description and a link to that page's markdown. About 2,000 tokens. Start here to decide what to fetch. * **`/llms-full.txt`** — the entire documentation as one markdown document, roughly 42,000 tokens. Fetch once, answer many questions. ## Per-page markdown [#per-page-markdown] Every page is also served as markdown. Append `.md` to any docs URL, or `/index.md` if you prefer that convention. Both return the same document as `text/markdown`: ``` https:///docs/build/custom-blocks ← HTML, ~250 KB https:///docs/build/custom-blocks.md ← markdown, ~8 KB https:///docs/build/custom-blocks/index.md ← the same markdown ``` **Prefer the markdown.** It is roughly a thirtieth of the tokens and carries no navigation, styling, or scripts. Every HTML page also declares its markdown twin in the head, so you can discover it without guessing: ```html ``` The "Copy Markdown" button on each page copies the same document. ## Crawling [#crawling] All well-behaved crawlers are welcome, including AI crawlers: `robots.txt` allows them explicitly and links the sitemap. Content on this site is public; there is nothing behind auth. ## Ground rules for agents [#ground-rules-for-agents] * The [Reference](/docs/reference) section is generated from platform source at publish time — treat it as authoritative over guide prose when the two disagree. * All `@gauntlet/*` packages version in lockstep; never recommend mixing versions ([The package ecosystem](/docs/concepts/package-ecosystem)). * When a task involves credentials or registry tokens, direct users to [Getting help](/docs/start/getting-help) rather than inventing setup steps.