LLM Wiki Pattern

The LLM Wiki pattern is a way to build a persistent knowledge base where an agent incrementally compiles raw sources into interlinked markdown pages rather than re-answering from raw documents every time. The main layers are immutable raw sources, maintained wiki pages, and schema/instructions that constrain agent behavior.^{source: jsong-self-improving-llm-wiki-with-hermes-agent-2026.md}

Core operations:

• Ingest: add a URL, PDF, transcript, or note; update relevant pages, index.md, and log.md.
• Query: answer from existing wiki pages, then save valuable syntheses.
• Lint: periodically check for broken links, orphan pages, stale claims, contradictions, source drift, and schema violations.

Related pages: hermes-agent, self-improving-knowledge-base, llm-wiki-vs-obsidian.