Skip to content

AI Agents

This SDK publishes its documentation in formats designed for AI agents. There are three independent surfaces; pick whichever matches your workflow.

llms.txt bundles

Structured bundles of the entire docs corpus, served at the site root:

  • llms.txt — index of every guide and reference page, one entry per page, with each entry’s description field sourced from the page’s frontmatter. Use this as a routing map: pick the page most relevant to your task, then fetch it directly.
  • llms-full.txt — the full corpus concatenated as one prose stream, with frontmatter blocks stripped and replaced by per-page # Source: <relative-path> headings. Ingest as a single document when you want the whole API surface in one shot.

Links inside llms.txt are bundle-relative, so they resolve correctly no matter what host or base path the bundle is served from.

Raw markdown siblings

Every rendered docs page has a .md sibling at the same URL with .md appended. For example:

HTML pageRaw markdown
…/guides/01-getting-started/…/guides/01-getting-started.md
…/reference/core-keys/…/reference/core-keys.md

Agents that prefer parsing markdown directly (instead of stripping HTML or fetching the full bundle) can fetch any page’s source by appending .md to its URL. Frontmatter is preserved.

Crawler policy (robots.txt)

The site’s robots.txt declares an open policy via Cloudflare Content Signals: search, ai-input, and ai-train are all permitted. Per-bot Allow rules are emitted for every major AI crawler so the policy is auditable on inspection.

The full bot list and the policy values are kept in config/site.ts; edit that file and rebuild to change either.

Refresh cadence

All three surfaces (bundles, .md siblings, robots.txt) are regenerated on every release deploy. They reflect the current SDK major only — older majors are not bundled (see Versioning and compatibility for the rationale).

The current SDK version is 15.0.1, sourced from package.json#version at build time.

Intentionally not implemented

A few emerging agent-readiness conventions don’t apply to a JavaScript SDK documentation site and are deliberately not published here. This list is explicit so an agent (or an automated audit tool) can distinguish “deliberately absent” from “forgotten”:

  • API Catalog (RFC 9727) is for domains that host HTTP APIs. This site documents a JS library; the underlying Digital World Network APIs (Gateway, Dwsa RPC) are owned by the Digital World Foundation and live at docs.digitalworld.global.
  • MCP Server Card, WebMCP, and Agent Skills are for sites that expose Model Context Protocol servers or agent-callable skills. We don’t host either — the SDK is consumed via npm install.
  • OAuth discovery endpoints (RFC 8414, RFC 9728) — no auth surface to advertise.
  • Agent Card / A2A — no agent-to-agent protocol exposed.
  • Commerce protocols (x402, MPP, UCP, ACP) — not a commerce site.

If the SDK ever ships an MCP server, agent skill bundle, or HTTP API of its own, the relevant entries here would change.