Once the wiki is large enough (e.g., ~100 articles and ~400K words), an LLM agent can handle complex queries against it, researching answers. Rather than relying on a complex RAG setup, the LLM can auto-maintain indexes and summaries and read related data at this scale