Documentation for agents

Agents should read the same reviewed source as people. This site does not keep a hidden “AI version” of Lazurio documentation.

Discovery surfaces

• /llms.txt is a compact map of canonical pages and summaries.
• /content-index.json is the versioned structured index. It includes stable document IDs, routes, audiences, review metadata, canonical URLs, public source paths, evidence references and Markdown content.
• /sitemap-index.xml is the web crawler map generated by the documentation build.
• Human-readable pages remain the canonical interpretation surface.

Stable-ID retrieval defaults to the English document. Agents can request a different published locale explicitly; canonical routes always select the exact localized page.

All artifacts are built from data/v2/docs in the public documentation repository. The index identifies the exact source commit and whether it came from a dirty local preview. Production deployment rejects a dirty or non-approved commit.

Retrieval guidance

1. Start with llms.txt to select a page.
2. Prefer a stable ID from content-index.json when persisting a reference; routes may be localized later.
3. Read sourceRefs and the matching public source records before repeating a trust-critical claim.
4. Distinguish documented provider facts from Lazurio’s assessment.
5. Report the page review date and source commit when freshness matters.
6. If a statement is absent or deployment-specific, say so instead of filling the gap from inference.

Future MCP server

A later read-only MCP server may expose discovery and retrieval over the same content index. It will not become a writer, private overlay or second content store. The current JSON contract is the compatibility seam for that work; the website does not depend on the future server.