LLM Wikis
An LLM Wiki is a durable knowledge layer that gives humans, search engines, and AI agents the same basic map: stable pages, explicit metadata, reviewed sources, descriptive links, and a clear boundary between current truth and archive evidence.
What It Solves
LLM work often fails when evidence is scattered across raw files, chat histories, source dumps, and generated chunks. A wiki turns that pile into a deliberate structure:
- concept pages for explanations
- site hubs for ownership and scope
- source records for provenance
- decision pages for durable choices
- operations pages for repeatable work
- glossary pages for shared terms
- migration pages for URL and archive cleanup
What It Is Not
An LLM Wiki is not a raw vector database, a dump of every file, or a list of mechanical chunks. Retrieval can use the wiki, but the wiki itself should be readable before an embedding model touches it.
Core Components
| Component | Role |
|---|---|
| Canonical pages | Stable public pages with descriptive slugs and self-contained summaries. |
| Source records | Provenance pages that preserve source path, hash, headings, and related topic pages. |
| Typed relationships | Links that say whether a page is a parent, source, evidence, related topic, decision, or migration target. |
| Discovery files | llms.txt, llms-full.txt, XML sitemaps, REST catalog, and graph API. |
| Review metadata | Owner role, status, visibility, indexing policy, last reviewed date, and review cycle. |