Andrej Karpathy's 'LLM Wiki' turns raw documents into dynamic personal knowledge bases

Andrej Karpathy, renowned AI researcher and former Tesla AI director, released a GitHub Gist in April 2026 outlining his 'LLM Wiki' concept. The idea proposes using large language models to create and maintain persistent, structured personal knowledge bases. Rather than repeatedly dumping raw documents into an LLM chat for ad-hoc queries, users would compile an evolving wiki that the AI helps organize, summarize, and link. This turns the LLM from a transient search tool into a living knowledge system that can be referenced, edited, and expanded over time.

Karpathy argues that current workflows waste context and require re-processing the same information. An LLM Wiki would store structured entries, automatically updated as new information is added. The system could use retrieval to surface relevant content, but the wiki itself becomes the primary memory. This approach aligns with trends in agentic AI and persistent memory, potentially revolutionizing how professionals manage research, notes, and learning. Karpathy's influence ensures this concept will spark serious discussion and experimentation in the AI community.