What Is an Agent Wiki?

An agent wiki is a knowledge base whose primary reader is an AI agent. The idea has public lineage — Andrej Karpathy's LLM-wiki musings, and agentwikis.com's compiled public wikis — and a refreshingly boring architecture:

• A wiki is just a folder of Markdown files with YAML frontmatter (title, type, status, last_verified, summary, sources) plus a small registry describing its scope. No database. Versionable with git, editable with anything, owned by you.
• Three serving surfaces from the same files:
• llms.txt / llms-full.txt — the emerging convention for telling any agent what a site knows;
• an MCP server — list, read, and search tools any agent framework can call;
• plain semantic HTML — humans read the same pages; no JavaScript required.

Why this form wins for corporate knowledge

Zero infrastructure to operate, no vendor lock-in (it's Markdown — every tool and every future agent can read it), auditable freshness (last_verified on every page is a lintable claim, not a hope), and the compile-once economics described in The Corporate Brain.

This entire site is an agent wiki, built with the same engine we install for clients. Point your agent at /llms.txt — or ask us for the MCP endpoint — and it can read everything you are reading now.