LLM-wiki pattern

A design pattern for building personal knowledge bases where an LLM agent incrementally maintains a persistent, interlinked wiki of markdown files derived from raw sources.

Defining moves

1. Don’t re-derive on every query (the RAG default) — compile once, then keep current.
2. Three layers: see three-layer-architecture.
3. Three operations: see ingest-query-lint.
4. Human role: curate sources, ask good questions, decide what matters.
5. LLM role: read, summarize, cross-reference, file, maintain consistency, suggest the next question.

Why this beats raw RAG

compounding-knowledge — every ingest improves the substrate the next query reads from. Cross-references already exist. Contradictions already flagged. Synthesis already reflects everything read.

Lineage

Conceptual ancestor: memex (Vannevar Bush, 1945). The missing piece in Bush’s vision — who maintains the trails between documents — is what the LLM now provides.

Instantiated in this vault as

• Schema: CLAUDE.md
• Index: index.md
• Log: log.md
• Wiki: wiki/{entities,concepts,sources,calendar,synthesis}/
• Raw: raw/