In this video, we build an agent to chat with our documents without any RAG, but using Andrej Karpathy’s idea of an LLM wiki, completely with local tools. This can be a strong alternative to RAG, where the LLM often has to rediscover knowledge from scratch on every question. The idea here is different. Instead of retrieving from raw documents at query time, the LLM uses an already optimized, searchable knowledge base.
We use Ollama’s gemma4 model as the LLM, LangChain to create our agent and provide it with tools and memory, Streamlit to create a chat UI, and Obsidian to view the generated markdown documents.
You can watch it here: https://youtu.be/4D8FjzJXJd4