Andrej Karpathy Unveils 'LLM Knowledge Bases' as a New Product Category, Replacing Vector Databases with AI-Maintained Markdown Wikis

Andrej Karpathy unveils 'LLM Knowledge Bases,' a groundbreaking system where AI acts as a full-time research librarian, compiling self-healing Markdown wikis that replace vector databases and RAG pipelines, sparking enterprise interest as a potential new product category.

• Andrej Karpathy introduces 'LLM Knowledge Bases,' a system where an AI acts as a full-time research librarian, compiling and maintaining interconnected Markdown files to preserve project context across sessions, bypassing the need for vector databases and RAG pipelines.
• The system operates in three stages — raw data ingest, LLM-driven compilation into a structured wiki with backlinks and summaries, and continuous 'linting' health checks where the AI self-heals inconsistencies — making knowledge fully human-readable and auditable.
• The approach is sparking enterprise interest, with developers envisioning multi-agent swarm architectures and businesses recognizing that unstructured internal data could be transformed into a living, self-updating 'Company Bible,' with Karpathy suggesting the methodology represents an entirely new product category.