Unlike standard RAG, which is “stateless” (forgetting everything after each query), an LLM Wiki is cumulative:
Compilation over Search: Knowledge is “compiled” once and kept up to date. When a new document is added, the LLM doesn’t just index it; it updates existing Wiki pages to reflect the new information.
Optimized for Long Context: Thanks to models like Gemini 1.5 Pro or Claude 3.5, you can load the entire Wiki (hundreds of thousands of words) directly into the model’s memory, allowing it to reason across the whole corpus rather than isolated snippets.
Transparency: You can open the Wiki in editors like Obsidian and visually see how ideas are connected, correcting the AI if it misses or misinterprets a connection.