What problem does it solve?
AI agents often re-derive knowledge because they lack a durable wiki layer between raw sources and answers, so insights vanish and contradictions go unnoticed.
llm-wiki keeps raw evidence untouched while having the agent synthesize summaries, concept articles, and answers into a structured markdown wiki that compounds knowledge over time.
Core Features & Use Cases
- Schema-first workflows: Always read the schema to discover raw, wiki, index, and log locations before touching files, preventing accidental tree creation.
- Ingest, query, and lint disciplines: Follow documented steps for ingesting sources, responding to questions with citations, and auditing contradictions, orphan pages, and gaps.
- Index and log maintenance: Keep the index as a navigation catalog, append-only logs with standardized headings, and record contradictions alongside every change.
- Use Case: In a research knowledge base, ingest a new article, generate linked topic pages, update the index/log, and answer future queries with wiki-backed citations instead of rerunning RAG.
Quick Start
Ingest the latest raw source, update the index and log per the schema, document any contradictions, and file a new wiki page with the synthesis.