LLM Wiki Personal Knowledge Base

Problem

Personal notes and source material often sit as scattered files. RAG re-searches raw documents for every question, producing answers assembled on the fly with little durable structure. Manually maintaining a wiki is burdensome because cross-links, conflict notes, and summaries need constant updating. llm-wiki shifts that maintenance work to the LLM, creating a persistent, accumulating Markdown knowledge base between raw files and user queries.

How it works

The core idea is incremental curation rather than plain indexing. It focuses on three operations:
- Ingest: stores raw files under raw/, reads index.md, extracts entities and concepts, creates or updates pages in wiki/, maintains [[page]] links, preserves source citations, and annotates conflicts.
- Query: locates relevant pages through index.md, synthesizes answers with citations, and saves valuable answers as new wiki pages when useful.
- Lint: detects contradictions, stale claims, orphan pages, missing concepts, and missing cross-references, then produces reports under health/.

Boundaries and notes

It fits personal knowledge, research notes, and project documentation that need ongoing synthesis. It is not a replacement for precise databases or real-time business search. Page conventions include the wiki/ directory, the #openclaw tag, source citations, and at least two related page links. Humans should curate sources, ask meaningful questions, and judge significance, while the LLM handles summarization, archiving, referencing, and consistency maintenance.